Papers
33
Total Citations
3,724
H-Index
23
About
Mrinal Kalakrishnan is a leading roboticist whose work spans the frontiers of deep reinforcement learning, manipulation, and legged locomotion. His research has fundamentally advanced how robots learn complex, real-world skills—from grasping objects to traversing rough terrain. He is perhaps best known for co-authoring "QT-Opt," a landmark 2018 paper with over 575 citations that introduced a scalable, vision-based deep RL approach for robotic grasping, demonstrating that robots could learn dynamic manipulation policies directly from camera inputs. His highly cited 2021 work, "How to train your robot with deep reinforcement learning" (536 citations), distills critical lessons from deploying RL on physical systems, bridging the gap between simulation and reality. Earlier in his career, Kalakrishnan made foundational contributions to quadruped locomotion, with papers like "Learning, planning, and control for quadruped locomotion over challenging terrain" (297 citations) and "Fast, robust quadruped locomotion over challenging terrain" (168 citations) establishing robust control architectures for robots like the LittleDog platform. His work on compliant manipulation and force control (163+ citations) further advanced safe, adaptive robot behavior. With over 2,600 total citations, Kalakrishnan’s research has shaped modern robotics, enabling machines to learn and act with unprecedented autonomy and robustness.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4Online movement adaptation based on previous sensor experiences223 citations · 2011
- 5Skill learning and task outcome prediction for manipulation207 citations · 2011
- 6Optimal distribution of contact forces with inverse-dynamics control194 citations · 2013
- 7Fast, robust quadruped locomotion over challenging terrain168 citations · 2010
- 8Learning force control policies for compliant manipulation163 citations · 2011
- 9Compliant quadruped locomotion over rough terrain154 citations · 2009
- 10