Excited to share that I’ll be joining UC San Diego for my PhD, advised by Professor Hao Su!

Many thanks to everyone who helped me along my research journey so far — I’m looking forward to continuing research in robot learning, manipulation, and simulation!

Accepted to ICLR 2025! :D

ManiSkill-HAB is my first first-author work, and it would not have been possible without the mentorship, guidance, and support of @stonet2000.bsky.social and Hao Su, and I'm incredibly thankful! I'm also thankful for the feedback provided by the Hillbot and Hao Su Lab teams.

🔓 Everything is open source!

• Paper: arxiv.org/abs/2412.13211
• Code: github.com/arth-shukla/mshab
• Models: huggingface.co/arth-shukla/mshab_checkpoints
• Datasets: arth-shukla.github.io/mshab/#dataset-section

We hope our environments, baselines, and dataset are useful to the community :)
(5/5)

📊 We're releasing a massive dataset and generation tools to help the community solve these tasks

• 466GB of RGBD + state data
• 44K episodes
• 8.8M transitions
• Detailed event labeling + trajectory filtering

Download: arth-shukla.github.io/mshab/#dataset-section
(4/5)

🤖 We provide extensive RL & IL baselines and model checkpoints for whole-body control, tackling complex, very long-horizon rearrangement tasks. Each task chains multiple skills (Pick, Place, Open, Close) with simultaneous navigation & manipulation. (3/5)

⚡️ MS-HAB provides a GPU-accelerated implementation of the Home Assistant Benchmark (HAB) with realistic low-level control for successful grasping, manipulation, & interaction, all while achieving 3x the speed of prior work at similar GPU memory usage. (2/5)

📢 Introducing ManiSkill-HAB: A benchmark for low-level manipulation in home rearrangement tasks!

- GPU-accelerated simulation
- Extensive RL/IL baselines
- Vision-based, whole-body control robot dataset

All open-sourced: arth-shukla.github.io/mshab
🧵(1/5)