Trajectory¶
In this task the aircraft needs to follow a configuration specified by the track controller. Reward is provided based on the \(L_{2}\) Euclidian distance from the ideal position at time \(\mathbf{s}_{t}\). The task is intended to investigate non-linear control problems.
Usage¶
env = gym.make("trajectory-v1")
Default Configuration¶
{
"observation": {
"type": "Dynamics"
},
"action": {
"type": "ContinuousAction"
},
"area": (1024, 1024), # terrain map area [tiles]
"vehicle_type": "Dynamic", # vehicle type, only dynamic available
"duration": 10.0, # simulation duration [s]
"collision_reward": -200.0, # max -ve reward for crashing
"traj_reward": 10.0, # max +ve reward for reaching for following trajectory
"normalize_reward": True, # whether to normalize the reward [-1, +1]
"trajectory_config": {
"name": "climb",
"final_height": 200.0,
"climb_angle": 10.0 * np.pi / 180.0,
"length": 15.0,
} # trajectory configuration details
}
Specifically this is defined in: