# rail-berkeley/rlkit

Collection of reinforcement learning algorithms

Repository: https://github.com/rail-berkeley/rlkit
Canonical: https://ross.abutalabs.com/products/rlkit
Language: Python
License: MIT
License Family: permissive
Last push: 2024-06-17T17:33:51+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 3143, "days_push": 807, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2932, forks 573 (observed 2026-08-28T04:07:30.697914+00:00)

## What it is
RLkit is a PyTorch-based reinforcement learning framework and algorithm collection from UC Berkeley RAIL. It provides reference implementations of algorithms such as SAC, TD3, DQN, HER, AWAC, IQL, and Skew-Fit with example scripts.

## Use cases
- implement soft actor-critic experiments in pytorch
- reproduce reinforcement learning research papers
- train goal-conditioned policies with hindsight experience replay
- run offline RL with implicit q-learning
- learn how RL algorithms like TD3 and DQN are implemented
- prototype new deep RL algorithms on top of an existing framework

## When to choose
- you need reference PyTorch implementations of classic and offline RL algorithms
- you are doing RL research and want readable, hackable algorithm code
- you want example scripts for gym-style environments

## When to avoid
- you need a production RL training service with distributed scaling
- you want a high-level API like Stable-Baselines3 with many maintained environments
- you need actively developed features or recent RL algorithm support

## Facets
- artifact type: library
- maturity: maintenance
- function: reinforcement-learning, machine-learning, benchmarking
- domain: reinforcement-learning, machine-learning, artificial-intelligence
- platform: python
- tags: pytorch, research, off-policy-learning, goal-conditioned-rl, offline-rl

## Member repositories
- rail-berkeley/rlkit (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:30.697914+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T07:33:23.667202+00:00, confidence not recorded.
  - readme: https://github.com/rail-berkeley/rlkit (fetched 2026-08-28T04:07:30.697914+00:00, sha 22a356aadf77)
- Data as of 2026-08-30T08:39:29.467469+00:00.