What’s In an Observation

An observation is returned by the gym environment after resetting and stepping. An observation is a dictionary with the following keys:

  • 'ego_idx': index of the ego agent in the list of agents

  • 'scans': list of length num_agents of numpy.ndarrays of (num_beams, ), each array is the corresponding laser scan of the agent in the list

  • 'poses_x': list of length num_agents of floats, each agent’s x pose in the world

  • 'poses_y': list of length num_agents of floats, each agent’s y pose in the world

  • 'poses_theta': list of length num_agents of floats, each agent’s theta pose in the world

  • 'linear_vels_x': list of length num_agents of floats, each agent’s current longitudinal velocity

  • 'linear_vels_y': list of length num_agents of zeros

  • 'ang_vels_z': list of length num_agents of floats, each agent’s current yaw rate

  • 'collisions': list of length num_agents of 1s or 0s, whether each agent is in collision with another agent or the environment