About the "done" mask from environment

I was using the “done” information from the environment as a part the observation. It means I did minor modification to the challenge.py file :
action = gent.act(state, done) #[it was action = gent.act(state)]

Can I pass the binary variable “done” to the agent? Or the agent should have self-awareness of when it reach the target?

Thank you,

Thanks for your question. You should not use the binary variable “done” as part of the observation. The agent does not need to have self-awareness that it has reached the goal. Once it converges to the goal, the episode will automatically finish. In this sense, passing “done” as part of the observation will not be helpful in terms of solving the task.

One a related note, are we able to modify challenge.py file in our submissions? Would the default challenge file not be used?

I am asking because I cannot find a way to get the Target Position to the Agent by the current framework of instantiating a new agent.

Hi @Ruddra_dev, the target position is given by the first two dimensions of obs['sensor']. It gives goal location in the agent’s frame. You don’t need to change challenge.py file.

1 Like

Thank you for the Answer. Would you also help me understand what the other two dimensions of obs[‘sensor’] correspond to?

Hi @Ruddra_dev, obs['sensor'][2] and obs['sensor'][3] correspond to linear velocity (in m/s) and angular velocity (rad/s) in robot base frame.

1 Like