Air Combat Environment for multi-agent RL

[ardian-selmonaj]

Hi Developers and Community,

I have modified the Skybolt source code to use it as the rendering engine for my multi-agent reinforcement learning (RL) air combat environment.

The main changes were made in PythonBindings.cpp, where I extended the Python bindings to expose additional functionality and support rendering of JSBSim-based simulated aircraft. To represent opposing teams, I color the aircraft red and blue. For the cannon fire mechanism, I use a simplified approach based on ShuttleSRB objects scaled down to a much smaller size. In the Python bindings, this looks as follows:

int fire_rocket(int entityId)
    {
        // changed Matrix component in ShuttleSRB.osgt file to decrease size of Rocket.
        
        sim::EntityPtr rocket = engineRoot->entityFactory->createEntity("ShuttleSRB");
        engineRoot->scenario->world.addEntity(rocket);

        auto it = aircraftMap.find(entityId);
        if (it == aircraftMap.end())
        {
            throw std::runtime_error("FlightSim: aircraft with ID " + std::to_string(entityId) + " not found");
        }
        
        setPosition(*rocket, *getPosition(*it->second));
        setOrientation(*rocket, *getOrientation(*it->second));
		setVelocity(*rocket, *getOrientation(*rocket) * Vector3(3000,0,0));
		
        rocketMap[rocket->getId().entityId] = { rocket, currentWallTime };

        return rocket->getId().entityId;
    }

I plan to continue working on this project later this year. For now, I would like to thank the developers for their impressive work and for making Skybolt available to the community.

P.S. If interested, you can find my Air Combat model here.

[matthew-reid]

Hi Ardian,

Thanks for sharing how you're using Skybolt! I’ll definitely take a look at the paper you linked.

Your approach to doing cannon fire looks like a practical solution as the Skybolt OSG based renderer lacks a first-class representation of short-lived visual effect objects. Creating a small "fire effect" entities is a sensible approach.

Matt