# nv-tlabs/ASE

Adversarial skill embeddings for training reusable controllers for physically simulated characters.

Repository: https://github.com/nv-tlabs/ASE
Canonical: https://ross.abutalabs.com/products/ase
Homepage: https://research.nvidia.com/labs/toronto-ai/ASE/
Language: Python
License: NOASSERTION
License Family: other
Topics: animation, physics-simulation, reinforcement-learning
Last push: 2025-12-07T09:02:20+00:00

## Health v2 (maintenance only)
Score: 57/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 56, release rhythm 35, longevity 100
- inputs: {"age_days": 1591, "days_push": 269, "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 1116, forks 152 (observed 2026-08-28T04:03:38.470367+00:00)

## What it is
ASE (Adversarial Skill Embeddings) is a research codebase from NVIDIA for training reusable low-level skill controllers for physically simulated characters using reinforcement learning and adversarial imitation on large unstructured motion datasets. It builds on Isaac Gym and supports pre-training skill embeddings plus task-training of high-level controllers on top of them.

## Use cases
- train reusable motor skill controllers for physically simulated humanoid characters
- imitate large datasets of unstructured motion clips without task-specific annotation
- train high-level task controllers on top of pre-trained skill embeddings
- simulate life-like character behaviors like running to targets and getting up after falling
- test robustness of learned controllers against perturbations like thrown projectiles

## When to choose
- you need the original reference implementation of the ASE paper for research reproduction
- you work with Isaac Gym and want massively parallel physics-based character skill learning
- you want to study adversarial skill embeddings or hierarchical RL for character animation

## When to avoid
- you want maintained, easy-to-use implementations - the repo is deprecated in favor of MimicKit
- you need production character animation tooling rather than research code
- you cannot obtain Isaac Gym, which is required and no longer actively distributed

## Facets
- artifact type: library
- maturity: abandoned
- function: machine-learning, reinforcement-learning, simulation, animation
- domain: machine-learning, reinforcement-learning, simulation, graphics
- platform: python, windows
- tags: physics-based-character-animation, adversarial-imitation-learning, isaac-gym, skill-embeddings, motion-imitation, deprecated, research-code, gpu, linux

## Member repositories
- nv-tlabs/ASE (main) score 57

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:38.470367+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 2026-08-30T06:41:59.220939+00:00, confidence not recorded.
  - readme: https://github.com/nv-tlabs/ASE (fetched 2026-08-28T04:03:38.470367+00:00, sha 966b57ebe174)
  - homepage: https://research.nvidia.com/labs/toronto-ai/ASE/ (fetched 2026-08-29T12:46:02.597867+00:00, sha 19c66b5c21e9)
- Data as of 2026-08-30T08:39:29.467469+00:00.