Open spacemanC opened 10 months ago
Are you referring to getting metrics for all agents in simulation? By default the rewards/metrics from the environment are computed for each agent (they will return an array of shape [num_objects]).
If you call env.metrics
or env.rewards
on the state in the multi-actors demo (https://github.com/waymo-research/waymax/blob/main/docs/notebooks/multi_actors_demo.ipynb), then you will be able to get results for each object in simulation.
Hi, thanks for this amazing effort. How can intelligent agents with different rules access their evaluation results? Is there a sample reference for this