About me

I'm a PhD student advised by Prof. Jeannette Bohg in the Stanford Interactive Perception and Robot Learning Lab. My research lies at the intersection of machine learning, symbolic AI, motion planning, and robot control. My goal is to bring us closer to a world where robots can assist us in our daily lives.

I am particularly interested in using Task and Motion Planning (TAMP) to solve long-horizon manipulation problems. Prior TAMP works have been limited to simulation due to their computational complexity and reliance on perfect perception, dynamics, and control. My work seeks to overcome these limitations so we can run TAMP on a real robot to perform tasks in the real world.

Publications

Task-Agnostic Policy Sequencing

Chris Agia*, Toki Migimatsu*, Jiajun Wu, and Jeannette Bohg.

arXiv preprint, 2022. [bib]

Active Task Randomization: Learning Visuomotor Skills for Sequential Manipulation by Proposing Feasible and Novel Tasks

Kuan Fang*, Toki Migimatsu*, Ajay Mandlekar, Li Fei-Fei, and Jeannette Bohg.

arXiv preprint, 2022. [bib]

Category-Independent Articulated Object Tracking with Factor Graphs

Nick Heppert, Toki Migimatsu, Brent Yi, Claire Chen, and Jeannette Bohg.

IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2022. [bib]

Symbolic State Estimation with Predicates for Contact-Rich Manipulation Tasks

Toki Migimatsu, Wenzhao Lian, Jeannette Bohg, and Stefan Schaal.

IEEE International Conference on Robotics and Automation (ICRA), 2022. [bib]

Grounding Predicates through Actions

Toki Migimatsu and Jeannette Bohg.

IEEE International Conference on Robotics and Automation (ICRA), 2022. [bib]

OmniHang: Learning to Hang Arbitrary Objects using Contact Point Correspondences and Neural Collision Estimation

Yifan You*, Lin Shao*, Toki Migimatsu, and Jeannette Bohg.

IEEE International Conference on Robotics and Automation (ICRA), 2021. [bib]

Concept2Robot: Learning Manipulation Concepts from Instructions and Human Demonstrations

Lin Shao, Toki Migimatsu, Qiang Zhang, Karen Yang, and Jeannette Bohg.

Robotics: Science and Systems (RSS), 2020. [bib]

Object-Centric Task and Motion Planning in Dynamic Environments

Toki Migimatsu and Jeannette Bohg.

IEEE Robotics and Automation Letters (RA-L) with ICRA option, 2020. [bib]

Learning to Scaffold the Development of Robotic Manipulation Skills

Lin Shao, Toki Migimatsu, and Jeannette Bohg.

IEEE International Conference on Robotics and Automation (ICRA), 2020. [bib]

Controlling Muscle-Actuated Articulated Bodies in Operational Space

Toki Migimatsu*, Samir Menon*, and Oussama Khatib.

18th International Symposium on Robotics Research (ISRR), 2017. [bib]

A Parameterized Family of Anatomically Accurate Human Upper-Body Musculoskeletal Models for Dynamic Simulation & Control

Samir Menon, Toki Migimatsu, and Oussama Khatib.

IEEE International Conference on Humanoid Robots (Humanoids), 2016. [bib]

Driverseat: Crowdstrapping Learning Tasks for Autonomous Driving

Pranav Rajpurkar, Toki Migimatsu, Jeff Kiske, Royce Cheng-Yue, Sameep Tandon, Tao Wang, and Andrew Ng.

32nd International Conference on Machine Learning (ICML), 2015. [bib]

An Empirical Evaluation of Deep Learning on Highway Driving

Brody Huval, Tao Wang, Sameep Tandon, Jeff Kiske, Will Song, Joel Pazhayampallil, Mykhaylo Andriluka, Pranav Rajpurkar, Toki Migimatsu, Royce Cheng-Yue, Fernando Mujica, Adam Coates, and Andrew Ng.

arXiv preprint, 2015. [bib]

Coding

I enjoy coding in my spare time, and I have developed a number of general-purpose robotics libraries over the years. I generally prototype in Python and code in C++ for performance-sensitive applications like robot control, but I'm a huge fan of Rust and Haskell.

spatial-dyn: A fast, lightweight robot dynamics library for C++/Python optimized for controlling robots using Operational Space control. It uses Featherstone's efficient spatial dynamics algorithms, and features a highly stable variational integrator based on discrete dynamics. It was used to teach Topics in Advanced Robotic Manipulation (CS326).
redis-gl: A modern web-based visualizer for SpatialDyn. It allows users to interact with and apply forces to the robot with the cursor, while providing a real-time monitor of system variables for debugging. Parts of it have been used for visualizing SO(3) representations, manipulator Jacobians, velocity/force duality, and inertia matrices in Introduction to Robotics (CS223a).
ctrl-utils: A header-only library of utilities useful for robot control, including an asynchronous message-passing framework based on Redis (like ROS, but runs on any OS). This is what we use to control all the robots and integrate all the perception sensors in the lab.
symbolic: A C++/Python library for symbolic planning using the standard PDDL specification.
franka-panda: A driver for running Operational Space control on the Franka Emika Panda with basic friction compensation.
dbot-redis: A Redis wrapper for the Dynamic Bayesian Object Tracker (DBOT), which can visually track objects from 3D point cloud data.

Also, check out my .vimrc and .tmux.conf setup here.

Teaching

Winter 2020: Principles of Robot Autonomy II (AA 274b): Course Assistant

Fall 2019: Principles of Robot Autonomy I (AA 274a): Course Assistant

SCPD teaching recognition

Winter 2018: Introduction to Robotics (CS 223a): Head Course Assistant

CS department award

Fall 2017: Safe and Interactive Robotics (CS 333): Course Assistant

Spring 2017: Experimental Robotics (CS 225a): Co-Head Course Assistant

Winter 2017: Introduction to Robotics (CS 223a): Course Assistant