Logs and Results#
Log Directories#
Default:
logs/<timestamp>-<config_name>/eval_embodiment.logManiSkill OOD batch mode:
logs/eval/<EVAL_NAME>/<timestamp>-<env_id>-<obj_set>/run_ppo.log
Terminal Metrics#
During evaluation, the terminal and logs report metrics such as:
eval/success_once— Task success rate (at least one success per episode)eval/return— Cumulative return
Video Output#
When env.eval.video_cfg.save_video: True, rollout videos are saved under:
<log_path>/video/eval/
log_path is determined by runner.logger.log_path and the timestamp directory passed by run_eval.sh.
TensorBoard#
When logger backends such as tensorboard are enabled in the config, detailed curves and statistics are available under runner.logger.log_path.