Krishnan Srinivasan

I am a research scientist at Toyota Research Institute and recently completed my PhD at Stanford University, advised by Professor Jeannette Bohg and Animesh Garg. My research focuses on enabling dexterous robotic manipulation through reinforcement learning, foundation models, and large-scale simulation. My current work is building towards dexterous long-horizon generalist policies through large-scale data collection and algorithmic approaches.

Email  /  CV  /  Google Scholar  /  GitHub  /  LinkedIn

Research Interests

My core research interests cover:

• Dexterous robotic manipulation
• Reinforcement learning for robotics
• Robot foundation models
• Long-horizon policy learning
• Large-scale simulation

Education

PhD, Computer Science — Stanford University — June 2025
Dissertation: Learning Dexterous Manipulation Policies for Tool-Use
Advisors: Jeannette Bohg, Animesh Garg

B.S., Computer Science & Mathematics — Yale University — May 2017
Thesis: Unsupervised Learning on ScRNASeq with Autoencoders
Advisor: Smita Krishnaswamy

Recent Publications

Recent work in dexterous manipulation, reinforcement learning, and foundation models for robotics.

A comprehensive examination of large behavior models applied to complex dexterous manipulation tasks across multiple domains.

A robust approach to behavior cloning that effectively learns from suboptimal demonstrations using robust world models.

A novel approach to visual multitask policy learning using asymmetric critic guided distillation for improved performance across diverse manipulation tasks.

A scalable approach to multi-finger grasp evaluation that enables robust transfer from simulation to real-world robotic systems.

An object-centric approach to task-space policy learning that excels at contact-rich dexterous manipulation scenarios.

An adaptive horizon actor-critic method designed for policy learning in contact-rich environments using differentiable simulators.

Borrowed from this website's template