Papers, notes & fieldwork
Writing
Robotics, learning, and the mathematics behind them.
2026
Humanoid SRL
An extra arm expands humanoid reach, support, and whole-body manipulation, with trajectory optimization and Learn-to-Teach control.
Spectral Skills: A Command Space for Humanoid Control
A latent command space for humanoid control, learned by predicting motion: tracking, chaining, and composing skills on a Unitree G1.
Taking Learn to Teach to Sea
Field notes and videos from running the L2T locomotion policy on Digit on Georgia Tech’s CAREN motion platform and aboard the Navy’s Stiletto, including runs underway at 10 and 22 knots.
Teaching a Humanoid to Walk on Shifting Ground
Granular contact physics and terrain-adaptive reinforcement learning for humanoid locomotion on shifting ground.
Stochastic Dominance and Risk-Aware Objectives
A compact note on zeroth-, first-, and second-order stochastic dominance, with figures, examples, and theorem-style summaries connecting dominance relations to monotone and risk-averse objectives.
2025
Distributional Inverse Reinforcement Learning
Recovering reward and return distributions from demonstrations for risk-aware imitation learning.
SEEC: Stable End-Effector Control
Model-enhanced residual learning for stable end-effector control during humanoid loco-manipulation.
RL-augmented Adaptive MPC
RL corrections to adaptive MPC so bipedal locomotion stays accurate on challenging terrain.
2024
Learn to Teach
One-stage privileged teacher-student learning for sample-efficient Digit locomotion over 12+ terrain and disturbance settings.
2023
Infer and Adapt
Learn locomotion rewards from demonstrations, then use those rewards to train bipedal policies that adapt to unfamiliar terrain.