RTT Lua Cookbook

Date:Aug 7, 2020



As of orocos toolchain-2.6 the deployment component launched by rttlua has been renamed from deployer to Deployer. This is to remove the differences between the classical deployer and rttlua and to facilitate portable deployment scripts. If you are using an orocos toolchain version prior to 2.6, use deployer instead.

What is this RTT-Lua stuff anyway?

Lua is a simple, small and efficient scripting language. The Lua RTT bindings provide access to most of the RTT API from the Lua language. Use-cases are:

  • writing deployment scripts in Lua
  • writing Lua components and services
  • using the rFSM Statecharts with RTT

To this end RTT-Lua consists of:

  • a Lua scriptable taskbrowser (rttlua-gnulinux etc. binaries)
  • a standard RTT Component which can be scripted with Lua
  • an RTT service which can extend existing components with Lua scripting

Most information here is valid for all three approaches. If not, this is explicitly mentioned. The listings are shown as interactively entered into the rttlua- REPL (read-eval-print loop), but could just the same be stored in a script file.

Getting started


Currently RTT-Lua is in OCL. Is is enabled by default but will only be built if the Lua-5.1 dependency (Debian: liblua5.1-0-dev, liblua5.1-0, lua5.1) is found.

CMake options:

  • BUILD_LUA_RTT: enable this to build the rttlua shell, the Lua component, and the Lua plugin.
  • BUILD_LUA_RTT_DYNAMIC_MODULES: (EXPERIMENTAL) build RTT and deployer as pure Lua plugins. Not recommended unless you know what you are doing.
  • BUILD_LUA_TESTCOMP: build a simple testcomponent that is used for testing the bindings. Not required for normal operation.

Setting up the path to rttlib

rttlib.lua is a Lua module, which is not strictly necessary, but highly recommended to load as it adds various syntactic shortcuts and pretty printing (Many examples on this page will not work without!). The easiest way to load it is to setup the LUA_PATH variable:

export LUA_PATH=";;$HOME/src/git/orocos/ocl/lua/modules/?.lua"

If you are a orocos_toolchain_ros user and do not want to hardcode the path like this, you can source the following script in your .bashrc:

RTTLUA_MODULES=`rospack find ocl`/lua/modules/?.lua
if [ "x$LUA_PATH" == "x" ]; then

Starting rttlua

$ ./rttlua-gnulinux
OROCOS RTTLua 1.0-beta3 / Lua 5.1.4 (gnulinux)

or for rtt_ros_integration users:

$ rosrun ocl rttlua-gnulinux
OROCOS RTTLua 1.0-beta3 / Lua 5.1.4 (gnulinux)

Now we have a Lua REPL that is enhanced with RTT specific functionality. In the following RTT-Lua code is indicated by a > prompt, while shell scripts are shown with the typical $.

Loading rttlib.lua

Before doing anything it is recommended to load rttlib. Like any Lua module this can be done with the require statement. For example:

$ ./rttlua-gnulinux
OROCOS RTTLua 1.0-beta3 / Lua 5.1.4 (gnulinux)
> require("rttlib")

As it is annoying having to type this each time, this loading can automated by putting it in the ~/.rttlua dot file. This (Lua) file is executed on startup of rttlua:


The (optional) last line enables colors.

Basic commands (read this!)

  • rttlib.stat() Print information about component instances and their state
> rttlib.stat()
Name                State               isActive  Period
lua                 PreOperational      true      0
Deployer            Stopped             true      0
  • rttlib.info() Print information about available components, types and services
> rttlib.info()
services:   marshalling scripting print LuaTLSF Lua os
typekits:   rtt-corba-types rtt-mqueue-transport rtt-types OCLTypekit
types:      ConnPolicy FlowStatus PropertyBag SendHandle SendStatus TaskContext array bool
            bools char double float int ints rt_string string strings uint void
comp types: OCL::ConsoleReporting OCL::FileReporting OCL::HMIConsoleOutput OCL::HelloWorld
            OCL::LuaComponent OCL::LuaTLSFComponent OCL::TcpReporting

Where’s my TaskContext?


> tc = rtt.getTC()

Above code calls the getTC() function, which returns the current TC and stores it in a variable tc. For showing the interface just write =tc. In the repl the equal sign is a shortcut for return, which in turn causes the variable to be printed. (BTW: This works for displaying any variable)

> =tc
TaskContext: lua
state: PreOperational
isActive: true
getPeriod: 0
peers: Deployer
  lua_string (string) =  // string of lua code to be executed during configureHook
  lua_file (string) =  // file with lua program to be executed during configuration
  bool exec_file(string const& filename) // load (and run) the given lua script
  bool exec_str(string const& lua-string) // evaluate the given string in the lua environment

Since (rttlua beta5) the above does not print the standard TaskContext operations anymore. To print these, use tc:show().

Getting persistent history with rlwrap

rttlua does not offer persistent history like in the taskbrowser. If you want it, you can use rlwrap and to wrap rttlua as follows:

alias rttlua='rlwrap -a -r -H ~/.rttlua-history rttlua-gnulinux'

If you run rttlua it should have persistent history.


The following shows the basic API, see section Automatically creating and cleaning up component interfaces for a more convenient way add/remove ports/properties.

Creating Ports

> pin = rtt.InputPort("string")
> pout = rtt.OutputPort("string")
> =pin
 [in, string, unconn, local] //
> =pout
 [out, string, unconn, local] //

Both In- and OutputPorts optionally take a second string argument (name) and third argument (description).

Connecting Ports


For this the ports don’t have to be added to the TaskContext:

> =pin:connect(pout)
> return pin
 [in, string, conn, local] //
> return pout
 [out, string, conn, local] //

Using the Deployer

The rttlua-* REPL automatically creates a deployment component that is a peer of the lua taskcontext:

> tc = rtt.getTC()
> depl = tc:getPeer("Deployer")
> cp=rtt.Variable("ConnPolicy")
> =cp
> depl:connect("compA.port1","compB.port2", cp)

RTT Types and Typekits

Which types are available?

> rttlib.info()
services:       marshalling, scripting, print, os, Lua
typekits:       rtt-types, rtt-mqueue-transport, OCLTypekit
types:          ConnPolicy, FlowStatus, PropertyBag, SendHandle, SendStatus, TaskContext,
                array, bool, bools, char, double, float, int, ints, rt_string, string, strings, uint, void
comp types:     OCL::ConsoleReporting, OCL::FileReporting, OCL::HMIConsoleOutput,
                OCL::HelloWorld, OCL::LuaComponent, OCL::TcpReporting, OCL::TimerComponent,
                OCL::logging::Appender, OCL::logging::FileAppender,
                OCL::logging::LoggingService, OCL::logging::OstreamAppender, TaskContext

Creating RTT types

> cp = rtt.Variable("ConnPolicy")
> =cp
> cp.data_size = 4711
> print(cp.data_size)

Accessing global RTT constants

Printing the available constants:

> =rtt.globals

Accessing constants - just index!

> =rtt.globals.LOCK_FREE

Convenient initalization of multi-field types

It is cumbersome to initalize complex types with many subfields:

> tc = rtt.getTC()
> depl = tc:getPeer("Deployer")
> depl:import("kdl_typekit")
> t=rtt.Variable("KDL.Frame")
> =t
> t.M.X_x=3
> t.M.Y_x=2
> t.M.Z_x=2.3

To avoid this, use the fromtab() method:

> t:fromtab({M={Z_y=1,Y_y=2,X_y=3,Y_z=4,Z_z=5,Y_x=6,Z_x=7,X_x=8,X_z=9},p={Y=3,X=3,Z=3}})

or even shorter using the table-call syntax of Lua,

> t:fromtab{M={Z_y=1,Y_y=2,X_y=3,Y_z=4,Z_z=5,Y_x=6,Z_x=7,X_x=8,X_z=9},p={Y=3,X=3,Z=3}}

Initalization of array/sequence types

When you created an RTT array type, the initial length will be zero. You must set the length of an array before you can assign elements to it (starting from toolchain-2.5 fromtab will do this automatically):

> ref=rtt.Variable("array")
> ref:resize(3)
> ref:fromtab{1,1,10}
> print(ref) -- prints {1,1,10}



> p1=rtt.Property("double", "p-gain", "Proportional controller gain")

(Note: the second and third argument (name and description) are optional and can also be set when adding the property to a TaskContext)

Adding to TaskContext Interface

> tc=rtt.getTC()
> tc:addProperty(p1)
> =tc -- check it is there...

Getting a Properties from a TaskContext

> tc=rtt.getTC()
> pgain = tc:getProperty("pgain")
> =pgain -- will print it

Properties of basic types: setting the value

> p1:set(3.14)
> =p1  -- a property can be printed!
p-gain (double) = 3.14 // Proportional controller gain

In particular, the following will not work:

> p1=3.14

Lua works with references! This will assign the variable p1 a numeric value of 3.14 and the reference to the property is lost.

Properties of basic types: getting the value

> print("the value of " .. p1:info().name .. " is: " .. p1:get())
the value of p-gain is: 3.14

Properties of complex types: accessing

Assume a property of type KDL::Frame. Similarily to Variables the subfields can be accessed by using the dot syntax:

> d = tc:getPeer("Deployer")
> d:import('kdl_typekit')
> f=rtt.Property('KDL.Frame')
> =f
(KDL.Frame) = {M={Z_y=0,Y_y=1,X_y=0,Y_z=0,Z_z=1,Y_x=0,Z_x=0,X_x=1,X_z=0},p={Y=0,X=0,Z=0}} //
> f.M.Y_y=3
> =f.M.Y_y
> f.p.Y=1
> =f
(KDL.Frame) = {M={Z_y=0,Y_y=3,X_y=0,Y_z=0,Z_z=1,Y_x=0,Z_x=0,X_x=1,X_z=0},p={Y=1,X=0,Z=0}} //

Like Variables, Properties feature a fromtab method to initalize a Property from values in a Lua table. See Section RTT Types and Typekits - Convenient initalization of multi-field types for details.


As properties are not automatically garbage collected, property memory must be managed manually:

> tc:removeProperty("p-gain")
> =tc         -- p-gain is gone now
> p1:delete() -- delete property and free memory
> =p1         -- p1 is 'dead' now.
userdata: 0x186f8c8


Synchronous calling of operations from Lua:

Calling Operations

The short and convenient way

> d = tc:getPeer("Deployer")
> =d:getPeriod()

The significantly faster and real-time safe way (because locally cached)

> d = tc:getPeer("Deployer")
> op = d:getOperation("getPeriod")
> =op -- can be printed!
double getPeriod() // Get the configured execution period. -1.0: no thread ...
> =op() -- call it

Sending Operations

“Sending” Operations permits to asynchronously request an operation to be executed and collect the results at a later point in time.

> d = tc:getPeer("Deployer")
> op = d:getOperation("getPeriod")
> handle=op:send() -- calling it
> =handle:collect()
SendSuccess    0


  • collect() returns multiple arguments: first a SendStatus string (SendSuccess, SendFailure) followed by zero to many output arguments of the operation.
    • collect blocks until the operation was executed, collectIfDone() will immediately return (but possibly with SendNotReady)
  • If your code make excessive use of “Sending Operations” something in your application design is probably wrong.

Can I define new Operations from Lua?

Answer: No.

Workaround: define a new TaskContext that inherits from LuaComponent and add the Operation there. Implement the necessary glue between C++ and Lua by hand (not hard, but some manual work required).


Loading and Using

For example, to load the marshalling service in a component and then to use it to write a property (cpf) file:

> tc=rtt.getTC()
> depl=tc:getPeer("Deployer")
> depl:loadService("lua", "marshalling") -- load the marshalling service in the lua component
> =tc:provides("marshalling"):writeProperties("props.cpf")

A second (and slightly faster) option is to get the Operation before calling it:

> -- get the writeProperties operation ...
> writeProps=tc:provides("marshalling"):getOperation("writeProperties")
> =writeProps("props.cpf") -- and call it to write the properties to a file.

What Operations and Ports are provided by a Service?

> depl:loadService("lua", "marshalling") -- load the marshalling service
> depl:loadService("lua", "scripting") -- load the scripting service
> print(tc:provides())
Service: lua
Subservices: marshalling, scripting
Operations:  activate, cleanup, configure, error, exec_file, exec_str, getPeriod,
                inFatalError, inRunTimeError, isActive, isConfigured, isRunning,
                setPeriod, start, stop, trigger, update
    Service: marshalling
    Operations:  loadProperties, readProperties, readProperty, storeProperties,
                    updateFile, updateProperties, writeProperties, writeProperty
    Service: scripting
    Operations:  activateStateMachine, deactivateStateMachine, eval, execute,
                    getProgramLine, getProgramList, getProgramStatus, getProgramStatusStr,
                    getProgramText, getStateMachineLine, getStateMachineList,
                    getStateMachineState, getStateMachineStatus, getStateMachineStatusStr,
                    getStateMachineText, hasProgram, hasStateMachine, inProgramError,
                    inStateMachineError, inStateMachineState, isProgramPaused, isProgramRunning,
                    isStateMachineActive, isStateMachinePaused, isStateMachineRunning,
                    loadProgramText, loadPrograms, loadStateMachineText, loadStateMachines,
                    pauseProgram, pauseStateMachine, requestStateMachineState, resetStateMachine,
                    runScript, startProgram, startStateMachine, stepProgram,
                    stopProgram, stopStateMachine, unloadProgram, unloadStateMachine

Accessing the Global Service

The RTT Global Service is useful for loading services into your application that don’t belong to a specific component. Your C++ code accesses this object by calling


The GlobalService object can be accessed in Lua using a call to:

gs = rtt.provides()

Which you can access later-on again using the rtt table:

rtt.provides("os"):argc() -- returns the number of arguments of this application
rtt.provides("os"):argv() -- returns a string array of arguments of this application


You can add different types of Activities to your component:

  • periodic activity
-- create activity for producer: period=1, priority=0,
-- schedtype=ORO_SCHED_OTHER (1).
depl:setActivity("producer", 1, 0, rtt.globals.ORO_SCHED_RT
  • non-periodic activity
-- create activity for producer: period=0, priority=0,
-- schedtype=ORO_SCHED_OTHER (1).
depl:setActivity("producer", 0, 0, rtt.globals.ORO_SCHED_RT)
  • master-slave activity:
    • Attach a (non-)periodic activity to the master component
    • Indicate that a component is the slave of a master
depl:setMasterSlaveActivity("name_of_master_component", "name_of_slave_component")

Basic usage patterns

How to write a deployment script

(see also the example in section How to write a RTT-Lua component)

-- deploy_app.lua

tc = rtt.getTC()
depl = tc:getPeer("Deployer")

-- import components, requires correctly setup RTT_COMPONENT_PATH
-- depl:import("componentX")
-- import components, requires correctly setup ROS_PACKAGE_PATH (>=Orocos 2.7)

-- create component 'hello'
depl:loadComponent("hello", "OCL::HelloWorld")

-- get reference to new peer
hello = depl:getPeer("hello")

-- create buffered connection of size 64
cp = rtt.Variable('ConnPolicy')
cp.type=1   -- type buffered
cp.size=64  -- buffer size
depl:connect("hello.the_results", "hello.the_buffer_port", cp)
rtt.logl('Info', "Deployment complete!")

run it:

$ rttlua-gnulinux -i deploy-app.lua

or using rtt_ros_integration

$ rosrun ocl rttlua-gnulinux -i deploy-app.lua


The -i option makes rttlua enter interactive mode (the REPL) after executing the script. Without it would exit after finishing executing the script, which in this case is probably not what you want.

How to write a RTT-Lua component

A Lua component is created by loading a Lua-script implementing zero or more TaskContext hooks in a OCL::LuaComponent. The following RTT hooks are currently supported:

  • bool configureHook()
  • bool activateHook()
  • bool startHook()
  • void updateHook()
  • void stopHook()
  • void cleanupHook()
  • void errorHook()

All hooks are optional, but if implemented they must return the correct return value (if not void of course). It is also important to declare them as global (by not adding a local keyword. Otherwise they would be garbage collected and not called)

The following code implements a simple consumer component with an event-triggered input port:


-- The Lua component starts its life in PreOperational, so
-- configureHook can be used to set stuff up.
function configureHook()
inport = rtt.InputPort("string", "inport")    -- global variable!
cnt = 0
return true

-- all hooks are optional!
--function startHook() return true end

function updateHook()
local fs, data = inport:read()
rtt.log("data received: " .. tostring(data) .. ", flowstatus: " .. fs)

-- Ports and properties are the only elements which are not
-- automatically cleaned up. This means this must be done manually for
-- long living components:
function cleanupHook()

A matching producer component is shown below:

require "rttlib"


function configureHook()
outport = rtt.OutputPort("string", "outport")    -- global variable!
cnt = 0
return true

function updateHook()
outport:write("message number " .. cnt)
cnt = cnt + 1

function cleanupHook()

A deployment script to deploy these two components:

require "rttlib"

depl = tc:getPeer("Deployer")

-- create LuaComponents
depl:loadComponent("producer", "OCL::LuaComponent")
depl:loadComponent("consumer", "OCL::LuaComponent")

--... and get references to them
producer = depl:getPeer("producer")
consumer = depl:getPeer("consumer")

-- load the Lua hooks

-- configure the components (so ports are created)

-- connect ports
depl:connect("producer.outport", "consumer.inport", rtt.Variable('ConnPolicy'))

-- create activity for producer: period=1, priority=0,
-- schedtype=ORO_SCHED_OTHER (1).
depl:setActivity("producer", 1, 0, rtt.globals.ORO_SCHED_RT)

-- raise loglevel

-- start components

-- uncomment to print interface printing (for debugging)
-- print(consumer)
-- print(producer)

-- sleep for 5 seconds
os.execute("sleep 5")

-- lower loglevel again


Automatically creating and cleaning up component interfaces

(available from toolchain-2.5)

The function rttlib.create_if can (re)generate a component interface from a specification as shown below. Conversely, rttlib.tc_cleanup will remove and destruct all ports and properties again.

-- stupid example:
iface_spec = {
        { name='inp', datatype='int', type='in+event', desc="incoming event port" },
        { name='msg', datatype='string', type='in', desc="incoming non-event messages" },
        { name='outp', datatype='int', type='out', desc="outgoing data port" },

        { name='inc', datatype='int', desc="this value is added to the incoming data each step" }

-- this create the interface

function configureHook()
    -- it is safe to be run twice, existing ports
    -- will be ignored. Thus, running cleanup() and configure()
    -- will reconstruct the interface again.

    inc = iface.props.inc:get()
    return true

function startHook()
    -- ports/props can be indexed as follows:
    return true

function updateHook()
    local fs, val
    fs, val = iface.ports.inp:read()
    if fs=='NewData' then iface.ports.outp:write(val+inc) end

function cleanupHook()
    -- remove all ports and properties

How to write a RTT-Lua Service

In contrast to Components (which typically contain functionality which is standalone), Services are useful for extending functionality of existing Components. The LuaService permits to execute arbitrary Lua programs in the context of a Componen

Simple example

The following dummy example loads the LuaService into a HelloWorld component and then runs a script that modifies a property:

require "rttlib"
d = tc:getPeer("Deployer")

-- create a HelloWorld component
d:loadComponent("hello", "OCL::HelloWorld")
hello = d:getPeer("hello")

-- load Lua service into the HelloWorld Component
d:loadService("hello", "Lua")

-- Execute the following Lua script (defined a multiline string) in
-- the service. This dummy examples simply modifies the Property.  For
-- large programs it might be better tostore the program in a separate
-- file and use the exec_file operation instead.
proggie = [[
    tc=rtt.getTC() -- this is the Hello Component
    prop = tc:getProperty("the_property")
    prop:set("hullo from the lua service!")

prop = hello:getProperty("the_property") -- get hello.the_property
print("the_property before service call:", prop)
hello:provides("Lua"):exec_str(proggie) -- execute program in the service
print("the_property after service call: ", prop)

Executing a LuaService function at the frequency of the host component

More useful than just running once is to be able to execute a function synchronously with the updateHook of the host component. This can be achieved by registering a ExecutionEngine hook (much easier than it sounds!).

The following Lua service code implements a simple monitor that tracks the currently active (TaskContext) state of the component in whose context it is running. When the state changes the new state is written to a port tc_state, which is added to the context TC.

This code could be useful for a supervision statemachine that can then easily react to this state change by means of an event triggered port.

require "rttlib"
d = tc:getPeer("Deployer")

-- create a HelloWorld component
d:loadComponent("hello", "OCL::HelloWorld")
hello = d:getPeer("hello")

-- load Lua service into the HelloWorld Component
d:loadService("hello", "Lua")

mon_state = [[
    -- service-eehook.lua
    tc=rtt.getTC() -- this is the Hello Component
    last_state = "not-running"
    out = rtt.OutputPort("string")
    tc:addPort(out, "tc_state", "currently active state of TaskContext")

    function check_state()
        local cur_state = tc:getState()
        if cur_state ~= last_state then
            last_state = cur_state
        return true -- returning false will disable EEHook

    -- register check_state function to be called periodically and
    -- enable it. Important: variables like eehook below or the
    -- function check_state which shall not be garbage-collected
    -- after the first run must be declared global (by not declaring
    -- them local with the local keyword)

-- execute the mon_state program


the -i option causes rttlua to go to interactive mode after executing the script (and not exiting afterwards).

$ rttlua-gnulinux -i service-eehook.lua
> rttlib.portstats(hello)
the_results (string)  =
the_buffer_port (string)  = NoData
tc_state (string)  = Running
> hello:error()
> rttlib.portstats(hello)
the_results (string)  =
the_buffer_port (string)  = NoData
tc_state (string)  = RunTimeError

How to perform runtime system validation?

It is often useful to validate a deployed system at runtime, however you want to avoid cluttering individual components with non-functional validation code. Here’s what to do (Please also see this post on orocos-users, which inspired the following)

Use-case: check for unconnected input ports

  1. Write a function to validate a single component

The following function accepts a TaskContext as an argument and checks whether it has unconnected input ports. If yes it prints an error.

function check_inport_conn(tc)
local portnames = tc:getPortNames()
local ret = true
for _,pn in ipairs(portnames) do
    local p = tc:getPort(pn)
    local info = p:info()
    if info.porttype == 'in' and info.connected == false then
        rtt.logl('Error', "InputPort " .. tc:getName() .. "." .. info.name .. " is unconnected!")
        ret = false
return ret
  1. After deployment, execute the validation function on all components:

This can be done using the mappeers function.

rttlib.mappeers(check_inport_conn, depl)

The mappeers function is a special variant of map which calls the function given as a first argument on all peers reachable from a TaskContext (given as a second argument). We pass the Deployer here, which typically knows all components.

Here’s a dummy deployment example to illustrate:

require "rttlib"

-- define or import check_inport_conn function here

-- dummy deployment, ports are left unconnected.
depl:loadComponent("hello1", "OCL::HelloWorld")
depl:loadComponent("hello2", "OCL::HelloWorld")

rttlib.mappeers(check_inport_conn, depl)

Executing it will print:

0.155 [ ERROR  ][/home/mk/bin//rttlua-gnulinux::main()] InputPort hello1.the_buffer_port is unconnected!
0.155 [ ERROR  ][/home/mk/bin//rttlua-gnulinux::main()] InputPort hello2.the_buffer_port is unconnected!

Using rFSM Statecharts with RTT

rFSM is a fast, lightweight Statechart implementation is pure Lua. Using RTT-Lua rFSM Statecharts can conveniently be used with RTT. The rFSM sources can be found here.

Where to run a statemachine: Component vs. Service?


Typically a Component will be preferred when

  • the statemachine has to coordinate/interact with/supervise multiple components
  • it shall run purely event-driven or at a different frequency than the computational components

A Service is preferred when

  • the Statemachine coordinates/monitors only one component
  • the Statemachine runs synchronous (same frequency) with the host component

There will, undoubtly, be exceptions!

How to run an rFSM in a Component

Summary: Create a OCL::LuaComponent. In configureHook load and initalize the fsm, in updateHook call rfsm.run(fsm)

(see the rFSM docs for general information)

It is a best-practice to split the initalization (setting up required functions, peers or ports used by the fsm) and the fsm model itself into two files. This way the fsm model is kept as platform independent and hence reusable as possible.

The following initalization file is executed in the newly create LuaComponent for preparing the environment for the statemachine, that is loaded and initalized in configureHook.


require "rttlib"
require "rfsm"
require "rfsm_rtt"
require "rfsmpp"

local tc=rtt.getTC();
local fsm
local fqn_out, events_in

function configureHook()
    -- load state machine
    fsm = rfsm.init(rfsm.load("fsm.lua"))

    -- enable state entry and exit dbg output
                    { STATE_ENTER=true, STATE_EXIT=true},

    -- redirect rFSM output to rtt log
    fsm.info=function(...) rtt.logl('Info', table.concat({...}, ' ')) end
    fsm.warn=function(...) rtt.logl('Warning', table.concat({...}, ' ')) end
    fsm.err=function(...) rtt.logl('Error', table.concat({...}, ' ')) end

    -- the following creates a string input port, adds it as a event
    -- driven port to the Taskcontext. The third line generates a
    -- getevents function which returns all data on the current port as
    -- events. This function is called by the rFSM core to check for
    -- new events.
    events_in = rtt.InputPort("string")
    tc:addEventPort(events_in, "events", "rFSM event input port")
    fsm.getevents = rfsm_rtt.gen_read_str_events(events_in)

    -- optional: create a string port to which the currently active
    -- state of the FSM will be written. gen_write_fqn generates a
    -- function suitable to be added to the rFSM step hook to do this.
    fqn_out = rtt.OutputPort("string")
    tc:addPort(fqn_out, "rFSM_cur_fqn", "current active rFSM state")
    rfsm.post_step_hook_add(fsm, rfsm_rtt.gen_write_fqn(fqn_out))
    return true

function updateHook() rfsm.run(fsm) end

function cleanupHook()
    -- cleanup the created ports.

A dummy statemachine stored in the fsm.lua file:

return rfsm.state {
    ping = rfsm.state {
        entry=function() print("in ping entry") end,

    pong = rfsm.state {
        entry=function() print("in pong entry") end,

    rfsm.trans {src="initial", tgt="ping" },
    rfsm.trans {src="ping", tgt="pong", events={"e_pong"}},
    rfsm.trans {src="pong", tgt="ping", events={"e_ping"}},

Option A: Running the rFSM example with a Lua deployment script


-- alternate lua deploy script
require "rttlib"


d:loadComponent("Supervisor", "OCL::LuaComponent")
sup = d:getPeer("Supervisor")

cmd = rttlib.port_clone_conn(sup:getPort("events"))

Run it. cmd is an inverse (output) port which is connected to the incoming (from POV of the fsm) events port of the fsm, so by writing to it we can send events:

$ rosrun ocl rttlua-gnulinux -i deploy.lua
OROCOS RTTLua 1.0-beta3 / Lua 5.1.4 (gnulinux)
INFO: created undeclared connector root.initial
> sup:start()
> in ping entry

> cmd:write("e_pong")
> in pong entry

> cmd:write("e_ping")
> in ping entry

> cmd:write("e_pong")
> in pong entry

Option B: Running the rFSM example with an Orocos deployment script


loadComponent("Supervisor", "OCL::LuaComponent")

After starting the supervisor we leave it, so we can write to the events ports:

$ rosrun ocl deployer-gnulinux -s deploy.ops
INFO: created undeclared connector root.initial
Switched to : Deployer

    This console reader allows you to browse and manipulate TaskContexts.
    You can type in an operation, expression, create or change variables.
    (type 'help' for instructions and 'ls' for context info)

        TAB completion and HISTORY is available ('bash' like)

Deployer [S]> cd Supervisor

TaskBrowser connects to all data ports of Supervisor
Switched to : Supervisor
Supervisor [S]> start
= true

Supervisor [R]> in ping entry

Supervisor [R]> leave
Watching Supervisor [R]> events.write ("e_pong")
= (void)

Watching Supervisor [R]> in pong entry

Watching Supervisor [R]> events.write ("e_ping")
= (void)

Watching Supervisor [R]> in ping entry

Watching Supervisor [R]>

Running rFSM in a Service

This is basically the same as executing a function periodally in a service (see the Service example above). There is a convenience function service_launch_rfsm in rfsm_rtt.lua to make this easier.

The steps are:

  • create LuaService in Component in question
  • prepare Lua environment, i.e. call exec_string or exec_file to add functions.
  • launch the fsm with the following call in your deployment script:
require "rfsm_rtt"

-- get reference to exec_str operation
fsmfile = "fsm.lua"
execstr_op = comp:provides("Lua"):getOperation("exec_str")
rfsm_rtt.service_launch_rfsm(fsmfile, execstr_op, true)

The last line means the following: launch fsm in <fsmfile> in service identified by execstr_op, true: create an execution engine hook so that the rfsm.step is called at the component frequency. (See the generated rfsm_rtt API docs).

Replacing states, functions and transitions of an existing FSM model

rFSM allows the creation of a FSM by loading a parent FSM into a new .lua file. This way, it is possible to add, delete and override states, transitions and functions. Though powerful, these operations can make the new FSM fairly hard to track. In this regard, a few tricks can make our life easier:

  • naming states and transitions in a consistent way
  • making the parent FSM as simple as possible with meaningful transition events
  • overriding a full state is less confusing than overriding a single entry or exit function

Generally speaking, the most effective way of creating a new FSM from a parent one is populating the original simple states by overriding them with composite states. In this context, the parent FSM provides “empty” boxes to be filled with application-specific code.

In the following example, “daughter_fsm.lua” loads “mother_fsm.lua” and overrides a state, two transitions and a function. “daughter_fsm.lua” is launched by a Lua Orocos component named “fsm_launcher.lua” . Deployment is done by “deploy.ops” . Instructions on how to run the example follow.



-- daughter_fsm.lua loads mother_fsm.lua
-- implementing extra states, transitions and functions
-- by adding and overriding the original ones.

require "utils"
require "rttros"

-- local variables to avoid verbose function calling
local state, trans, conn = rfsm.state, rfsm.trans, rfsm.conn

-- path to the fsm to load
local base_fsm_file = "mother_fsm.lua"

-- load the original fsm to override
local fsm_model=rfsm.load(base_fsm_file)

-- set colored outputs indicating the current state
dbg = rfsmpp.gen_dbgcolor( {STATE_ENTER=true}, false)

-- Overriding StateA
-- In "mother_fsm.lua" StateA is an rfsm.simple_state
-- Here we make it an rfsm.composite_state
fsm_model.StateA = rfsm.state {

        StateA1= rfsm.state {
                entry=function() print("in State A1") end,

        StateA2 = rfsm.state {
                entry=function() print("in State A2") end,

        rfsm.transition {src="initial", tgt="StateA1"},
        tr_A1_A2 = rfsm.transition {src ="StateA1", tgt="StateA2", events={"e_move_to_A2"}},
        tr_A2_A1 = rfsm.transition {src ="StateA2", tgt="StateA1", events={"e_move_to_A1"}},

-- Overriding single transitions
fsm_model.tr_A_to_B = rfsm.trans {src="StateA", tgt="StateB", events={"e_daughter_A_to_B"}}
fsm_model.tr_B_to_A = rfsm.trans {src="StateB", tgt="StateA", events={"e_daughter_B_to_A"}}

-- Overriding a specific function
fsm_model.StateB.entry = function()
                print("I am in State B in the daughter FSM")
return fsm_model


require "rttlib"
require "rfsm"
require "rfsm_rtt"
require "rfsmpp"

local tc=rtt.getTC();
local fsm
local fqn_out, events_in

function configureHook()
   -- load state machine
   fsm = rfsm.init(rfsm.load("daughter_fsm.lua"))

   -- enable state entry and exit dbg output
   fsm.dbg=rfsmpp.gen_dbgcolor("FSM loading example",
                   { STATE_ENTER=true, STATE_EXIT=true},

   -- redirect rFSM output to rtt log
   fsm.info=function(...) rtt.logl('Info', table.concat({...}, ' ')) end
   fsm.warn=function(...) rtt.logl('Warning', table.concat({...}, ' ')) end
   fsm.err=function(...) rtt.logl('Error', table.concat({...}, ' ')) end

   -- the following creates a string input port, adds it as a event
   -- driven port to the Taskcontext. The third line generates a
   -- getevents function which returns all data on the current port as
   -- events. This function is called by the rFSM core to check for
   -- new events.
   events_in = rtt.InputPort("string")
   tc:addEventPort(events_in, "events", "rFSM event input port")
   fsm.getevents = rfsm_rtt.gen_read_str_events(events_in)

   -- optional: create a string port to which the currently active
   -- state of the FSM will be written. gen_write_fqn generates a
   -- function suitable to be added to the rFSM step hook to do this.
   fqn_out = rtt.OutputPort("string")
   tc:addPort(fqn_out, "rFSM_cur_fqn", "current active rFSM state")
   rfsm.post_step_hook_add(fsm, rfsm_rtt.gen_write_fqn(fqn_out))
   return true

function updateHook() rfsm.run(fsm) end

function cleanupHook()
   -- cleanup the created ports.


loadComponent("Supervisor", "OCL::LuaComponent")

To test this example, run the Deployer:

rosrun ocl deployer-gnulinux -lerror -s deploy.ops


Deployer [S]> cd Supervisor

TaskBrowser connects to all data ports of Supervisor
   Switched to : Supervisor
Supervisor [R]> leave

Watching Supervisor [R]> events.write ("e_move_to_A2")

FSM loading example:    STATE_EXIT          root.StateA.StateA1
in State A2
FSM loading example:    STATE_ENTER         root.StateA.StateA2

One-liner to build a table of peers

A Coordinator often needs to interact with many or all other components in its vicinity. To avoid having to write peer1 = depl:getPeer("peer1") all over, you can use the following function to generate a table of peers which are reachable from a certain component (commonly the deployer):

peertab = rttlib.mappeers(function (tc) return tc end, depl)

Assume the Deployer has two peers “robot” and “controller”, they can be accessed as follows:

-- or


Connecting RTT Ports to ROS topics

> cp=rtt.Variable("ConnPolicy")
> cp.transport=3 -- 3 is ROS
> cp.name_id="/l_cart_twist/command" -- topic name
> depl:stream("CompX.portY", cp)

or with sweet one-liner (thx to Ruben!):

> depl:stream("CompX.portY", rtt.provides("ros"):topic("/l_cart_twist/command"))

Finding the path to a ROS package

This is sometimes usefull for loading scripts etc. that are located in different packages.

The rttros.lua collects some basic but useful stuff for interacting with ROS. This one is “borrowed” from the excellent roslua:

> require "rttros"
> =rttros.find_rospack("geometry_msgs")

How are types converted between RTT and Lua?

bool boolean
float number
double number
uint number
int number
char string
string string
void nil

This conversion is done in both directions: basic values read from ports or basic return values of operation are converted to Lua; vice versa if an operation with basic Lua values is called these will automatically be converted to the corresponding RTT types.

How to add a custom pretty printing function for a new type?

In short: write a function which accepts a lua table representation of you data type and returns either a table or a string. Assign it to rttlib.var_pp.mytype, where mytype is the value returned by the var:getType() method. That’s all!

Quick example: ConnPolicy type

(This is just an example. It has been done for this type already).

The out-of-box printing of a ConnPolicy will look as follows:

Orocos RTTLua 1.0-beta3 (gnulinux)
> return rtt.Variable("ConnPolicy")

This not too bad, but we would like to display the string representation of the C++ enums type and lock_policy. So we must write a function that returns a table…

function ConnPolicy2tab(cp)
    if cp.type == 0 then cp.type = "DATA"
    elseif cp.type == 1 then cp.type = "BUFFER"
    else cp.type = tostring(cp.type) .. " (invalid!)" end

    if cp.lock_policy == 0 then cp.lock_policy = "UNSYNC"
    elseif cp.lock_policy == 1 then cp.lock_policy = "LOCKED"
    elseif cp.lock_policy == 2 then cp.lock_policy = "LOCK_FREE"
    else cp.lock_policy = tostring(cp.lock_policy) .. " (invalid!)" end
    return cp

and add it to the rttlib.var_pp` table of Variable formatters as follows:

rttlib.var_pp.ConnPolicy = ConnPolicy2tab

now printing a ConnPolicy again calls our function and prints the desired fields:

> return rtt.Variable("ConnPolicy")

How to use classical OCL Deployers ? (like with Corba, or with a Taskbrowser)

If you are used to manage your application with the classic OCL Taskbrowser or if you want your application to be connected via Corba, you may only use lua for deployment, and continue to use your former deployer. To do so, you have to load the lua service into your favorite deployer (deployer, cdeployer, deployer-corba, …) and then call your deployment script.

Exemple : launch your prefered deployer :

cdeployer -s loadLua.ops

with loadLua.ops :

//load the lua service
loadService ("Deployer","Lua")

//execute your deployment file

and with yourLuaDeploymentFile.lua containing the kind of stuff described in this Cookbook. Like the one in paragraph “How to write a deployment script”

How to generate graphical representations of rFSM models

The rfsm-viz command allows you to generate easy-to-read pictures representing the structure of your FSM model. This tool uses the rfsm2uml and fsm2dbg modules and requires the libgv-lua package. Practically:

$ <fsm_install_dir>/tools/rfsm-viz -f <your_fsm_file>.lua


  • -f <fsm-file> : fsm input file
  • -tree : generate tree representation
  • -text : dump to simple textual format
  • -uml : generate uml state machine figure
  • -dot : generate a graphviz dot-file
  • -all : generate all represesentations
  • -format (svg|png|…): generate different file format
  • -v : be verbose
  • -h : print this

Script to generate default CPF file for a component

#!/usr/bin/env rttlua

if #arg~=3 or arg[1]=="--help" or arg[1] == "-h" then
   print [[Usage:
         bootstrap-cpf.lua package type output.cpf
           package: the package to import which contains the component
           type: the type of the component
           output.cpf: name of the property file

require 'rttlib'
if not dp:import(arg[1]) then
if not dp:loadComponent("comp",arg[2]) then
comp = dp:getPeer("comp")

Memory management: what is automatically garbage collected?

Answer: everything besides Ports and Properties. So if you have Lua components/Services which are deleted and recreated, it is advisable to cleanup properly. This means:

  • remove Port or Property from (all!) TaskContext interfaces to which it was added
  • invoke the delete method to release the memory e.g. portX:delete()

Update for toolchain-2.5: The utility function rttlib.tc_cleanup() will do this for you.