# amazon-far/holosoma

Repository: https://github.com/amazon-far/holosoma
Canonical: https://ross.abutalabs.com/products/holosoma
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-08-24T19:44:35+00:00

## Health v2 (maintenance only)
Score: 61/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 35, longevity 20
- inputs: {"age_days": 293, "days_push": 9, "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 1616, forks 249 (observed 2026-08-28T04:05:11.902992+00:00)

## What it is
Holosoma is a Python framework for training and deploying reinforcement learning policies on humanoid robots, supporting locomotion and whole-body tracking tasks across multiple simulators like IsaacGym, IsaacSim, MJWarp, and MuJoCo. It also includes motion retargeting to convert human motion capture data into robot motions and a shared inference pipeline for sim-to-sim and sim-to-real deployment.

## Use cases
- train RL locomotion policies for humanoid robots
- train whole-body tracking policies from motion capture
- retarget human mocap data to robot motions
- deploy trained policies to real humanoid robots
- evaluate policies in MuJoCo simulation
- log training videos and checkpoints with Wandb

## When to choose
- you are training RL policies for Unitree G1 or Booster T1 humanoids
- you need multi-simulator support (IsaacGym, IsaacSim, MJWarp, MuJoCo)
- you want an end-to-end pipeline from mocap retargeting to real-robot deployment
- you want PPO or FastSAC with shared sim/real inference

## When to avoid
- you work with non-humanoid robots
- you need Windows or macOS support
- you only need simple robot simulation without RL training
- you cannot access NVIDIA Isaac simulators or GPUs

## Facets
- artifact type: framework
- maturity: active
- function: reinforcement-learning, simulation, robotics, machine-learning, llm-training
- domain: robotics, reinforcement-learning, simulation, machine-learning
- platform: python
- tags: humanoid-robots, motion-retargeting, sim-to-real, isaacgym, mujoco, ppo, fastsac, unitree-g1, whole-body-tracking, locomotion, linux, gpu

## Member repositories
- amazon-far/holosoma (main) score 61

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:11.902992+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 2026-08-30T03:50:02.955549+00:00, confidence not recorded.
  - readme: https://github.com/amazon-far/holosoma (fetched 2026-08-28T04:05:11.902992+00:00, sha 425ed9bfeb6b)
- Data as of 2026-08-30T08:39:29.467469+00:00.