Python Layer

[stonier]

This is a meta-issue identifying the tasks required for Milestone #2. Some more detailed notetaking here.

Scenarios as a python script which is responsible for instantiating and configuring the simulator, agents, callback and evaluation functions, as well as ultimately starting and stepping through the simulation.

Simulator

Be able to instantiate the simulator with options so that the same script can be used in both interactive and CI modes.

• Instantiate the simulator
• able to start the simulation in paused mode
• able to start the simulation in 'as fast as possible' mode (i.e. non-realtime)
• able to run the simulation for a finite duration
• command-line options framework for pause, non-realtime, duration modes

Road Network

• instantiate a dragway (proof of life) [Issue Reimplement the dragway #270]
• instantiate a multilane roadway (where we want to be)

Agents

• add a simple car
• add a lane following car
• add a lane changing & merging (MOBIL) car

Dynamic Scripting

• execution time statistic collection, dump to stdout at the end of the simulation [Simulation statistics #333, Add real time rate to simulation stats #373]
• vehicle collision notification and abort out of the simulation loop, exit with failure (moved to Collision Detection #372)

Infra

• Simulation Runner should not be compiled as STATIC [Simulation Runner still STATIC #341]
• Switch to python3, Switch to using python3 #158 (moved to Collision Detection #372)

Documentation

• User and Developer notes in the Delphyne Guide. Document the caveats in the developer's section (i.e. new type, expose it in python_bindings.cc)

Demo Goal

• Multilane world with lane changing cars and finite amount of time
• One parked car (external), one simple car that collides into the parked car, aborts simulation (moved to Collision Detection #372)

---

Priorities

From least to highest

• Vehicle collisions : might be some work if we go the c++ route, also pending what robotics does
• External cars : since it's blocking on Chris' work anyway
• Remaining : should all be unblocked and not needing any design

Risks

• The python bindings bucket since it's very beta and we may need many parameter bindings
• Not yet finalised how we transport maliput and view it in the visualisers

[stonier]

@basicNew I just worked through this with @clalancette. Please rant/improve it with your recommendations :)

[basicNew]

Weekly update:

Got a first working version of realtime factor configuration in the core (c++) classes and python bindings. There is an issue though with dynamically changing it while the sim is running (see video). Some options:

• Do not allow dynamic reconfiguration.
• Allow it, but leave it as is.
• Raise the issue and, if considered a bug, send a fix to Drake.
• Handle the sim loop delay on the simulation runner (i.e. on the delphyne side).

[basicNew]

Weekly update:

• Created Issue on Drake about freezing / speeding simulation.
• Real time rate configuration added to core C++ classes and python bindings. Dynamic realtime rate changes are supported (demo video). [Better real-time rate support #254].
• Update keyboard controlled simulation to use simulation loop + callback instead of its external sim loop. Also fix issue with Play/Pause buttons due to issue discussed in first item [Change keyboard_controlled_simulation demo implementation #240].

[basicNew]

Weekly update:

All the items under Simulator are covered. Below is a description of the demos showing them:

• Instantiate the simulator. All the python demos do this. python_bindings_example.py shows the basics of instantiating a simulator with a callback on each simulation step.
• Able to start the simulation in paused mode:
  • paused_mode_example.py shows how to start the simulator paused.
  • python_bindings_example.py takes an optional command line flag to start the simulation paused.
  • keyboard_controlled_simulation.py reads the terminal keyboard while the simulation is running and can toggle play/pause.
• Able to start the simulation in 'as fast as possible' mode (i.e. non-realtime):
  • time_bounded_simulation.py can take the real-time factor as an optional command-line parameter.
  • realtime_rate_changer.py shows how to change dynamically the real-time factor while the simulation is running.
• Able to run the simulation for a finite duration:
  • time_bounded_simulation.py can take the the time to run the simulation as an optional command-line parameter (otherwise it just runs for 15 secs) and quit.
• Command-line options framework for pause, non-realtime, duration modes. Done as depicted above.

Currently working on the dragway demo. Have an initial implementation working, including a python example.

[stonier]

@basicNew one item we should add here is the STATIC simulator runner issue, #341
(relevant since it looks like a python gil problem).

And thanks! That helped get me up to speed very quickly, especially w.r.t. demo_launcher and the growing library of python examples.

[basicNew]

+1, will do. And yw, happy to help :).

[stonier]

Looking good. Transferred remaining jobs to the Final M2 meta: #378.