rlworkgroup/garage

Add option to Evaluator for recording videos

Open

#1,862 opened on Aug 6, 2020

View on GitHub
 (1 comment) (1 reaction) (0 assignees)Python (1,728 stars) (297 forks)batch import
featuregood first issue

Description

This is definitely a nice to have -- It'd be nice if as a part of snapshots, we could have the option to record the policy being rolled out deterministically for a few rollouts, so that we can watch its behavior in our various environments. I haven't had much success using the out of the box tool that open ai gym provides for this, called the gym Monitor wrapper. It requires that all environments internally return a Done signal after max_path_length steps. Perhaps we can get around this by wrapping environments with the open ai gymtimelimit wrapper.

Alternatively someone could look at the stable baselines vec video recorder wrapper. It seems that they have some type of video recorder, however it doesn't work out of the box with garage since stable baselines has pinned versions of TF and other libraries necessary to garage.

Contributor guide

Tech stack
python
Domain
machine learning
Issue type
feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
3-5 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
knowledge of garage and OpenAI Gymunderstanding of video recording wrappers
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
40
Research direction
Investigate the Gym Monitor wrapper and the Stable Baselines VecVideoRecorder wrapper to understand their implementation. Then design a similar wrapper for garage that can record videos of policy rollouts. The implementation should handle environments that do not return a done signal after max path length steps, possibly by using the TimeLimit wrapper. Look at the garage codebase for the Evaluator class and snapshot functionality to integrate the video recording option.