robocasa 开源项目推荐（17 个）

Tavish9 /any4lerobot

🎁 A collection of utilities for LeRobot.

zqc3117 /Spotter

Official codebase for Spotter: Let the Embodied Model Lead, and the VLM Reflect for It

xxxxyu /cosmos-lite

Efficient inference / rollout for Cosmos 3 Nano / Edge robot policies

Chaoqi-LIU /praxis-eval

A simple, efficient, and extensible evaluation framework for robot policies across diverse benchmarks, supporting both local and remote execution

robodreamer /embodied-policy-lab

Run and compare robot VLAs and world-action models in LIBERO and RoboCasa with reproducible local evidence.

aws-samples /sample-rfm-simulation-on-aws

1-click robot foundation model (VLA, world action model, LLM agent) simulation and evaluation on AWS GPU instances with AWS CDK

Chaoqi-LIU /sim-env

Example usage of a set of simulation benchmarks

DAVIAN-Robotics /PointmapVLA

PointmapVLA: a vision-language-action model that sees in metric 3D via robot-base-frame pointmaps. Code, checkpoints and fine-tuning/eval for LIBERO, LIBERO-Plus and RoboCasa.

pm1255 /robot_stack

Paired perturbation and recovery datasets for robot learning, with native simulator evidence, replay and audit tools.

only-one-over /robocasa-control

Chat-driven RoboCasa simulation control plugin for DeepSeek Harness

humma-humma /persistent-3d-memory-vla

Confidence-aware world-aligned 3D scene memory and SmolVLA integration for RoboCasa

jmberhold /dabtroll

VIBTARVE: VLM-initiated behavior-tree auditing of robot VLA execution. Common-evidence replay, multi-model evaluation, and human-reviewed RoboCasa GR1 tabletop recordings from GR00T N1.6.

Ethan-Chen-plus /radeon-physical-ai-evidence-suite

AMD ROCm Physical AI training, evaluation, simulation, rendering, and reproducible evidence suite

zhiyuan-gao /flf-latent-world-model

Action-conditioned latent world model and experiment code for RoboCasa365.

S1rlvk /smolwam-robocasa

SmolWAM (0.88B world-action model) on RoboCasa: zero-shot 0/30, fine-tuned 9/20 in held-out kitchens

LiXiling /mops-data

MOPS Data Generation

AMD-AGI /faster-action-models

Faster inference for World Action Models (FastWAM, DiT4DiT) on AMD GPUs: GEAK optimized kernels, fewer denoising steps, and Compact Action Experts.