As per discussion in chat w/ @thowell , when querying the best trajectory from an agent running CEM, the state field is all 0s. A rollout for the elite_avg trajectory needs to be computed every planner iteration and we should either remove the resetting of elite_avg or place it elsewhere relative to the nominal rollout.
As per discussion in chat w/ @thowell , when querying the best trajectory from an agent running CEM, the
state
field is all 0s. A rollout for theelite_avg
trajectory needs to be computed every planner iteration and we should either remove the resetting ofelite_avg
or place it elsewhere relative to the nominal rollout.