Research Areas
We develop algorithms and systems for robots that physically interact with complex real-world environments.
Physical Embodiment
We integrate force-aware control and tactile sensing to enhance the physical capabilities of robotic systems. By processing high-fidelity contact feedback and haptic data, our robots achieve safe and stable interaction in unstructured environments, moving beyond purely vision-based approaches.
Visuomotor Policy
We build end-to-end learning systems that map sensory inputs directly to motor commands. Utilizing imitation learning and Vision-Language-Action (VLA) models, our research enables robots to generalize across diverse tasks and follow complex natural language instructions in real-world settings.
Whole-Body Control
We develop optimization-based control frameworks for high-degree-of-freedom robotic systems. By leveraging Hierarchical Quadratic Programming (HQP) for multi-objective constraint satisfaction and Model Predictive Path Integral (MPPI) for nonlinear dynamics, we enable agile, balanced, and coordinated full-body movements.
Research Projects
Selected robot-learning and control demonstrations from PIER Lab.
View all videosFeatured Demonstrations
Recent demonstrations with available videos.
Quantizing Diffusion Policy
The demonstration for our CoRL 2026 paper, which quantizes diffusion policies through dynamic scaling and reweighted distillation.
Humanoid Bimanual Manipulation
ACT-based imitation learning driving bimanual manipulation on our IGRIS-C platform.
Articulate Object task and Object Placement
An ACT-based Franka FR3 manipulation demonstration combining door opening with object placement.
Latent-Action VLA
Latent-action-space flow-matching VLA.
On-Device Policy Inference
Running a learned manipulation policy within the compute and memory budget of embedded hardware — a hand-over task inferring on an NVIDIA Jetson Orin.
Humanoid Perceptive Locomotion
An RL-based Perceptive locomotion demonstration of the Unitree G1 humanoid walking on stairs.
Whole-Body Teleoperation
A whole-body teleoperation demonstration with reinforcement learning and Motion retargeting.
Self-Collision Avoidance in Whole-Body Control
Keeping a high-degree-of-freedom robot clear of self-collision while it tracks its task objectives with MPPI.
Indoor Navigation with MPPI
Model Predictive Path Integral control for indoor navigation, demonstrated on the Unitree G1 humanoid.
Dynamic Obstacle Avoidance
Reliable avoidance of moving obstacles during navigation with MPPI.