Top Robotics & Physical AI Researchers (2026)

Professor (EECS), UC Berkeley · Amazon Scholar; co-leads Frontier AI & Robotics, heads AGI LLM efforts, Amazon

PhD · Stanford University — Andrew Ng

Berkeley professor of robotics and deep RL, BAIR co-director, and Covariant co-founder. A pioneer of robot learning.

He pioneered deep RL and imitation learning and co-founded Covariant. Recent work includes DreamDojo, a general-purpose robot world model trained on 44,000 hours of human video. He also works on humanoid parkour with real robots.

Co-founder & CEO, Robot Era · Assistant Professor, Tsinghua University (IIIS)

PhD · UC Berkeley — Masayoshi Tomizuka (Mechanical Systems Control Lab)

Founder and CEO of Robot Era, which builds STAR1, XHand, and ERA-42. Tsinghua IIIS researcher in RL and contact-rich manipulation.

Robot Era’s founder is a faculty member at Tsinghua IIIS. He leads RL-based humanoid locomotion and transfer from simulation to real systems. He also works on VLA architectures, including BagelVLA and VLM4VLA.

Assistant Professor, Stanford University · Co-founder, Physical Intelligence

PhD · UC Berkeley — Pieter Abbeel & Sergey Levine (BAIR)

Pioneered meta-learning (MAML) and robot imitation learning. Stanford professor and co-founder of Physical Intelligence.

She pioneered meta-learning and co-founded Physical Intelligence. She coauthored pi-0.6, which uses RL self-improvement for VLAs that fold laundry and make espresso in real homes. She also coauthored Cosmos Policy, which fine-tunes a video model into a state-of-the-art robot policy.

Co-founder & CEO, Physical Intelligence

PhD · University of Southern California (USC) — Gaurav Sukhatme (Robotic Embedded Systems Lab)

Co-founder and CEO of Physical Intelligence and a robot-learning researcher. Previously at Google DeepMind. Stanford adjunct.

As CEO and co-founder of Physical Intelligence, he builds the pi-series VLA foundation models. Commercial deployments include laundry folding and packaging. He has long led work on robot learning across different bodies.

Associate Professor (EECS), UC Berkeley · Co-founder, Physical Intelligence

PhD · Stanford University — Vladlen Koltun (advisor); co-advised by Patrick Hanrahan

Studies deep RL and robot learning. Co-founded Physical Intelligence and contributes to pi0 and RT-2 robot foundation models.

The most-cited researcher in modern robot learning. He drove the shift to large-scale robot control based on data and co-founded Physical Intelligence. Recent work includes steerable and asynchronous VLA policies for embodied reasoning.

Co-founder & CEO, Skild AI · Raj Reddy Associate Professor of Robotics, Carnegie Mellon University

PhD · UC Berkeley — Alexei (Alyosha) Efros & Trevor Darrell (BAIR)

Skild AI co-founder and CEO and CMU robotics professor. Studies curiosity-driven exploration and transfer of robot learning from simulation to real systems.

Skild AI’s CEO originated curiosity-driven learning. He drives omni-bodied learning from human video and object-centric world models for general manipulation. This work includes Latent Particle World Models.

Tenure-track Assistant Professor, Peking University (CFCS / EPIC Lab) · Founder & CTO, Galbot

PhD · Stanford University — Leonidas J. Guibas (Geometric Computing Group)

At PKU CFCS and EPIC Lab, he studies embodied AI and 3D vision. Known for category-level 6D pose (NOCS) and dexterous grasping. Galbot founder and CTO.

He leads embodied AI and dexterous manipulation at Peking University. He built LDA-1B, a billion-parameter robot foundation model trained on 30,000 hours of embodied data.

Co-founder & CEO, Sunday Robotics

PhD · Stanford University (incomplete — left PhD in 2024) — Chelsea Finn (IRIS Lab); also part-time at Google DeepMind

Created ALOHA and ACT for low-cost bimanual robot learning. Co-founder and CEO of Sunday Robotics, which builds the Memo home robot.

Sunday Robotics’ CEO is first author of ACT and ALOHA. This low-cost system combines bimanual teleoperation and imitation learning. It became a standard for dexterous manipulation research.

Co-founder & CTO, Sunday Robotics

PhD · Columbia University — Shuran Song

Co-founder and CTO of Sunday Robotics, which builds the Memo home robot. Created Diffusion Policy and the UMI gripper.

Sunday Robotics’ CTO is first author of Diffusion Policy and UMI (Universal Manipulation Interface). These are two of the most widely adopted methods in robot imitation learning.

Tenured Associate Professor, School of CS / Center on Frontiers of Computing Studies; leads PKU-AgiBot Lab, Peking University

PhD · Imperial College London — Yike Guo

Peking University professor who studies embodied AI and robotic manipulation that generalizes. Leads PKU-AgiBot Lab.

Professor of Computer Science & Engineering, University of Washington · Senior Director of Robotics Research, Allen Institute for AI (Ai2)

PhD · University of Bonn — Computer Science (Armin B. Cremers' group)

Studies probabilistic robotics, particle filters, grasping, and pose estimation. Now leads robot foundation models at Ai2.

He leads robotics research at NVIDIA and UW. Recent work includes PointWorld’s 3D point-flow world models and scaling test-time computation for VLAs.

Co-founder & President, Skild AI · Professor, Robotics Institute, Carnegie Mellon University

PhD · University of Maryland, College Park — Larry Davis

CMU robotics professor and Skild AI co-founder and president. Studies self-supervised learning and large-scale robot learning.

Co-founder, Physical Intelligence

PhD · Stanford University — Marco Pavone (Autonomous Systems Lab), Aeronautics & Astronautics

Co-founded Physical Intelligence. Previously at Google Brain and DeepMind. Studies LLM-grounded robotics, including SayCan and Code as Policies.

Arthur J. Chick Professor of EECS, UC Berkeley · VP & Distinguished Scientist, Amazon FAR (Frontier AI & Robotics)

PhD · Stanford University — Thomas Binford's robotics/vision group

Computer-vision pioneer known for normalized cuts, R-CNN, and shape contexts. Now leads robotics research at Amazon FAR.

One of the most influential computer-vision researchers. He now drives vision for robotics and general-purpose robot world models at UC Berkeley and NVIDIA.

Lead Research Scientist, Robot Learning & Foundation Models, Microsoft Research

PhD · University of Freiburg — Wolfram Burgard

Studies robot learning, internet-scale foundation models, and VLA models, including Octo and CALVIN.

Member of Technical Staff, Physical Intelligence

PhD · University of Southern California (USC) — Joseph Lim (CLVR Lab)

Studies robot foundation models and large-scale datasets. Co-led Open X-Embodiment, OpenVLA, DROID, and Octo.

A lead contributor to OpenVLA, DROID, and FAST action tokenization. These open-source tools are central to VLA models that operate across robot bodies.

Roboticist, Meta Superintelligence Labs · Assistant Professor of Computer Science, New York University

PhD · Carnegie Mellon University — Abhinav Gupta

Studies robot learning, large-scale self-supervised grasping, and affordable open-source robots. Co-founded ARI, which Meta acquired.

He leads dexterous-manipulation research at NYU and advances self-supervised and large-scale robot learning.

Associate Professor, Khoury College of Computer Sciences, Northeastern University

PhD · University of Massachusetts Amherst — Autonomous Learning Laboratory (Andrew Barto / Roderic Grupen)

Northeastern roboticist who studies equivariant models for robot manipulation that needs less data. Also affiliated with RAI Institute.

Assistant Professor of Electrical Engineering (and Computer Science, by courtesy); Director, Robotics and Embodied AI Lab (REAL), Stanford University

PhD · Princeton University — Thomas Funkhouser (Princeton Vision & Robotics Group)

Leads Stanford REAL Lab. Co-created Diffusion Policy and UMI for visuomotor robot learning.

She leads 3D perception and manipulation research at Stanford. She contributes extensively to learned manipulation and scene understanding.

Toyota Professor, EECS / MechE / AeroAstro; member of CSAIL, Massachusetts Institute of Technology · Senior Vice President of Large Behavior Models, Toyota Research Institute

PhD · Massachusetts Institute of Technology — MIT (Computational Cognitive Science / Sebastian Seung)

MIT professor and TRI Senior Vice President of Large Behavior Models. Created Drake. Studies large behavior models for dexterous manipulation.

He leads robot control and underactuated-systems research at MIT and Toyota Research Institute. He created Drake simulation and the LBM/Diffusion-Policy line of manipulation research.

Cheryl and John Neerhout, Jr. Distinguished Professor of Mechanical Engineering, UC Berkeley

PhD · Massachusetts Institute of Technology — Mechanical Engineering (control systems)

Control theorist who directs the Mechanical Systems Control (MSC) Lab. Studies motion and robot control, mechatronics, and autonomous driving.

Professor, Department of Automation; Director, Intelligent Robotics and Machine Vision (IRMV) Lab, Shanghai Jiao Tong University

PhD · Chinese University of Hong Kong — Department of Automation & Computer-Aided Engineering, CUHK

Directs SJTU IRMV Lab. Studies deep learning for LiDAR odometry, visual servoing, and embodied AI robotics.

Co-founder & Chief Scientist, Generalist AI

PhD · Princeton University — Thomas Funkhouser

Former Google DeepMind roboticist and Generalist AI co-founder. Contributed to Transporter Networks, SayCan, and Code as Policies.

Associate Professor of Computer Science, University of Texas at Austin · Director & Distinguished Research Scientist (co-leads GEAR), NVIDIA Research

PhD · Stanford University — Fei-Fei Li & Silvio Savarese (Stanford Vision Lab)

UT Austin professor and NVIDIA GEAR co-lead. Studies general-purpose robot learning. Created robosuite and Eureka.

He leads robot learning and simulation at UT Austin and NVIDIA. He coauthored DreamZero, a 14-billion-parameter video-diffusion world model for real-time robot control with feedback.

Research Scientist, Physical Intelligence · PhD student, UC Berkeley

PhD · UC Berkeley — RAIL / BAIR (Sergey Levine)

Lead author of the pi0 (π0) VLA flow model. Co-led Octo, a general-purpose robot policy.

First author of pi-0, Physical Intelligence’s flagship VLA foundation model.

PhD student, Stanford University

Built Mobile ALOHA and HumanPlus. Studies legged locomotion through learning and whole-body humanoid manipulation.

First author of HumanPlus (CoRL 2024), which lets humanoids learn by following human motion.

Researcher / Engineer, RobotEra

PhD · Tsinghua University — Jianyu Chen (Shanghai Qi Zhi Institute / Tsinghua)

RobotEra researcher behind Humanoid-Gym and humanoid locomotion through transfer from simulation to real systems (DWL).

First author of Denoising World Model research for humanoid locomotion, which won RSS 2024 Outstanding Paper. A co-founder at RobotEra.

Co-founder & CTO, CollectedAI · PhD candidate, Stanford University

PhD · Stanford University — IRIS Lab (Chelsea Finn)

Lead author of DROID, a large-scale dataset of robot manipulation in real-world settings. CollectedAI co-founder and CTO.

First author of DROID (RSS 2024), the largest dataset of robot manipulation in real-world settings.

PhD student (Finn/Liang labs), Stanford University

Lead author of OpenVLA, an open-source VLA model for robot control.

First author of OpenVLA. This open-source, 7-billion-parameter VLA model became a community standard.

Founding Researcher / Member of Technical Staff, Physical Intelligence

PhD · Stanford University — Chelsea Finn & Silvio Savarese (Stanford AI Lab / IRIS)

Founding researcher at Physical Intelligence. Scales robot learning through VLA foundation models, including R3M and pi0.

A key contributor to Physical Intelligence’s R3M and pi-0 foundation models for visual representation and robot control.

Postdoctoral Researcher, UC Berkeley (BAIR)

PhD · New York University (Courant Institute) — Lerrel Pinto (NYU Robot Learning Lab)

Studies robot learning from demonstrations, including Behavior Transformers and Dobb-E for household manipulation.

Known for DynaMem, Behavior Transformers. Received a CoRL workshop best-paper award and contributes extensively to robot learning.

Research Scientist, Physical Intelligence · PhD student, UC Berkeley

PhD · UC Berkeley — RAIL / BAIR (Sergey Levine)

Berkeley robot-learning researcher. Works on FAST action tokenization for high-frequency, general-purpose VLA policies.

Co-first author of FAST. Its action-tokenization scheme made autoregressive VLAs much faster to train.