Arthur Allshire

About

Arthur Allshire is a robotics researcher specializing in deep reinforcement learning, dexterous manipulation, and GPU-accelerated simulation for robot learning. His work sits at the intersection of simulation infrastructure and sim-to-real transfer, tackling one of robotics' most persistent challenges: training agile, generalizable manipulation policies that hold up in the physical world. Allshire is a key contributor to **Isaac Gym**, NVIDIA's landmark GPU-based physics simulation platform (322 citations), which revolutionized robot learning by enabling massively parallel training entirely on GPU, dramatically accelerating policy development. Building on this foundation, his **DeXtreme** work (88 citations) demonstrated that deep RL could produce agile, human-level in-hand manipulation skills transferable to real robot hands — a significant milestone in dexterous robotics. His **DexPBT** research further scaled these capabilities to complex hand-arm systems using population-based training. Allshire has also contributed to reproducible benchmarking through shared remote robot platforms, latent action space learning, and symmetry-aware policy design. More recently, he helped introduce **MuJoCo Playground**, an open-source framework democratizing sim-to-real research. Across his career, his cumulative citation impact reflects work that has materially shaped how the robotics community approaches learning-based manipulation at scale.

Research Focus

Key Achievements

Top Papers

1. 1
   Isaac Gym: High Performance GPU-Based Physics Simulation For Robot\n Learning

   322 citations · 2021
2. 2
   DeXtreme: Transfer of Agile In-hand Manipulation from Simulation to Reality

   88 citations · 2023
3. 3
   Transferring Dexterous Manipulation from GPU Simulation to a Remote Real-World TriFinger

   45 citations · 2022
4. 4
   DexPBT: Scaling up Dexterous Manipulation for Hand-Arm Systems with Population Based Training

   22 citations · 2023
5. 5
   Symmetry Considerations for Learning Task Symmetric Robot Policies

   15 citations · 2024
6. 6
   Transferring Dexterous Manipulation from GPU Simulation to a Remote Real-World TriFinger

   15 citations · 2021
7. 7
   LASER: Learning a Latent Action Space for Efficient Reinforcement Learning

   6 citations · 2021
8. 8
   DeXtreme: Transfer of Agile In-hand Manipulation from Simulation to Reality

   5 citations · 2022
9. 9
   Demonstrating MuJoCo Playground

   3 citations · 2025
10. 10
    A Robot Cluster for Reproducible Research in Dexterous Manipulation

    3 citations · 2021

Key Collaborators

Contact & Links