Gabriel Margolis

I am a PhD student at the Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT, where I work on Embodied Intelligence. Advised by Pulkit Agrawal, I study learned control as a component of complete robotic systems. Previously, I received my BS ('20) and MEng ('21) degrees at MIT.

My goal is to advance reliable and helpful mobile robots for assistance, delivery, emergency response, and sport. To this end, I think that sim-to-real reinforcement learning is a promising tool for synthesizing unscriptable motor skills.

Email / Google Scholar / Twitter / GitHub / LinkedIn

Updates

Apr '23	Our sim-to-real soccer ball manipulation project, DribbleBot, received some cool coverage from MIT News and other news outlets. Check out our project video and find us at ICRA next month!
Dec '22	Our open-source codebase for sim2real transfer of continuously parameterized structured gaits to Unitree Go1 is now released, accompanying our CoRL 2022 paper. Try it out here.
Dec '22	I co-organized the Workshop on Sim-to-Real Robot Learning: Locomotion and Beyond at the 2022 Conference on Robot Learning. Thanks to all who participated! The full recording is now available here.
Jun '22	I presented our work on teaching robots to run fast at the Robotics: Science and Systems conference in New York City. At the conference banquet, our robot hit the dance floor.

Research

	Learning to See Physical Properties with Active Sensing Motor Policies Gabriel B. Margolis, Xiang Fu, Yandong Ji, Pulkit Agrawal Conference on Robot Learning, 2023 paper / project page Learn to see how terrains feel by collecting self-supervised data with information-maximizing motor skills.
	DribbleBot: Dynamic Legged Manipulation in the Wild Yandong Ji, Gabriel B. Margolis , Pulkit Agrawal International Conference on Robotics and Automation (ICRA), 2023 paper / project page Dynamic object manipulation with the legs, using onboard computation and sensing.
	Walk These Ways: Tuning Robot Control for Generalization with Multiplicity of Behavior Gabriel B. Margolis, Pulkit Agrawal Conference on Robot Learning, 2022 (Oral Presentation) paper / project page / code One learned policy embodies parameterized dynamic behaviors useful for different tasks.
	Rapid Locomotion via Reinforcement Learning Gabriel B. Margolis, Ge Yang Kartik Paigwar, Tao Chen, Pulkit Agrawal Robotics: Science and Systems, 2022 paper / project page / code High-speed running and spinning on diverse terrains with a single neural network.
	Learning to Jump from Pixels Gabriel B. Margolis, Tao Chen, Kartik Paigwar, Xiang Fu, Donghyun Kim, Sangbae Kim, Pulkit Agrawal Conference on Robot Learning, 2021 paper / project page A hierarchical control framework for dynamic vision-aware locomotion.
	Learning Robust Terrain-Aware Locomotion Gabriel B. Margolis Master's Thesis, 2021 thesis My Master's thesis. This work received the Ernst A. Guillemin Master’s Thesis Award in Artificial Intelligence and Decision Making.