# rail-berkeley/softlearning

Softlearning is a reinforcement learning framework for training maximum entropy policies in continuous domains. Includes the official implementation of the Soft Actor-Critic algorithm.

Repository: https://github.com/rail-berkeley/softlearning
Canonical: https://ross.abutalabs.com/products/softlearning
Homepage: https://sites.google.com/view/sac-and-applications
Language: Python
License: NOASSERTION
License Family: other
Topics: reinforcement-learning, soft-actor-critic, deep-learning, deep-reinforcement-learning, deep-neural-networks, machine-learning
Last push: 2023-11-29T14:16:25+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 2830, "days_push": 1008, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1437, forks 249 (observed 2026-08-28T04:04:43.759983+00:00)

## What it is
Softlearning is a deep reinforcement learning toolbox for training maximum entropy policies in continuous domains, and the official implementation of the Soft Actor-Critic (SAC) algorithm. It is built on TensorFlow/Keras with Ray Tune for experiment orchestration and distributed cloud training.

## Use cases
- train soft actor-critic agents on continuous control tasks
- run maximum entropy reinforcement learning experiments
- reproduce official SAC research results
- launch distributed RL training on cloud services with Ray
- train RL policies for robotic locomotion and manipulation
- benchmark off-policy deep RL algorithms on MuJoCo environments

## When to choose
- you need the canonical, research-grade SAC implementation
- you work with continuous-domain control and MuJoCo environments
- you want Ray-based scaling of RL experiments to cloud infrastructure
- you prefer TensorFlow/Keras for RL model definitions

## When to avoid
- you want a PyTorch implementation - use rlkit or other SAC ports instead
- you need actively maintained software with recent updates
- you want RL for discrete action spaces
- you cannot obtain a MuJoCo license or prefer license-free simulators

## Facets
- artifact type: library
- maturity: maintenance
- function: reinforcement-learning, machine-learning, deep-learning, cli
- domain: reinforcement-learning, machine-learning, robotics, deep-learning
- platform: python
- tags: soft-actor-critic, maximum-entropy-rl, tensorflow, ray-tune, mujoco, continuous-control, off-policy, linux, docker, gpu

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

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:43.759983+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 2026-08-30T04:36:45.139921+00:00, confidence not recorded.
  - readme: https://github.com/rail-berkeley/softlearning (fetched 2026-08-28T04:04:43.759983+00:00, sha 18087d476c27)
  - homepage: https://sites.google.com/view/sac-and-applications (fetched 2026-08-29T11:47:36.021166+00:00, sha ce199e9abf64)
- Data as of 2026-08-30T08:39:29.467469+00:00.