Nathan Lambert
Meta (United States), University of California, Berkeley, Meta (Israel)
Papers
6
Total Citations
78
H-Index
4
About
Nathan Lambert is a robotics researcher whose work sits at the intersection of reinforcement learning, model-based control, and multi-agent systems. His core contributions focus on making robots more adaptable and sample-efficient in real-world settings. Lambert’s most influential work, “Learning Generalizable Locomotion Skills with Hierarchical Reinforcement Learning” (42 citations), introduces a framework that dramatically improves how legged robots learn to navigate diverse terrains and reach arbitrary goals on physical hardware. He has also advanced the field of model-based reinforcement learning, notably investigating and mitigating compounding prediction errors in learned dynamics models—a critical challenge for long-horizon robotic planning. In multi-robot systems, Lambert developed BotNet, a simulator that studies how realistic communication models (including latency and packet loss) affect swarm control and decentralized coordination. His work on low-level quadrotor control using deep model-based RL further demonstrates his commitment to bridging simulation and hardware reality. With over 78 total citations across his top papers, Lambert is recognized for tackling fundamental bottlenecks in robot learning—from hierarchical skill acquisition to accurate long-term prediction—making his research essential reading for anyone interested in deploying intelligent, adaptive robots in the physical world.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4
- 5
- 6Investigating Compounding Prediction Errors in Learned Dynamics Models4 citations · 2022