In preparation, 2027
Extending physics-embedded RL from recovery-as-goal to balance-as-infrastructure: a single policy that walks, handles perturbations, exploits environmental contacts (walls, tables) to extend stability regions, and carries objects through recovery — without dropping them.
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2026), 2026
We show that injecting classical balance metrics — capture point, center-of-mass state, and centroidal momentum — as privileged critic inputs and physics-grounded reward terms enables a single RL policy to recover from any initial pose (supine, seated, kneeling, crouching) with 93.4% success, using emergent multi-contact strategies (elbows, knees, forearms) without prescribed contact schedules.
Recommended citation: Poddar, N. (2026). Embedding Classical Balance Control Principles in Reinforcement Learning for Humanoid Recovery. IROS 2026. https://arxiv.org/abs/2603.08619
AIAA SCITECH 2026 Forum, 2026
A learned model prunes the search space for classical path planners (A*, RRT), predicting which regions are unlikely to contain optimal paths and focusing computation where it matters — yielding significant speedups with minimal solution quality degradation.
Recommended citation: Poddar, N., Mishra, B., Clark, G., Sevil, H. E., & Griffin, R. (2026). Accelerating Classical Path Planning via Learned Search Space Reduction. AIAA SCITECH 2026 Forum. https://doi.org/10.2514/6.2026-1997 https://arc.aiaa.org/doi/abs/10.2514/6.2026-1997
2025 IEEE-RAS 24th International Conference on Humanoid Robots (Humanoids), 2025
Rather than replicating operator foot poses directly, we retarget user steps to robot footstep locations and predict upcoming steps to reduce timing delays — enabling seamless teleoperation synchronization on uneven terrain on the humanoid robot Nadia.
Recommended citation: Penco, L., Park, B., Fasano, S., Poddar, N., McCrory, S., Kitchel, N., Bialek, T., Anderson, D., Calvert, D., & Griffin, R. (2025). Anticipatory and Adaptive Footstep Streaming for Teleoperated Bipedal Robots. IEEE-RAS Humanoids 2025. https://arxiv.org/abs/2508.11802