Spring 2025 GRASP SFI: Student Lightning Talks, Session 2

This will be a hybrid event with in-person attendance in Levine 307 and virtual attendance on Zoom.

ABSTRACT

The Spring 2025 GRASP SFI Student Lightning Talks will highlight the research of three GRASP Lab Master’s or early PhD students whose presentation topics have been nominated by their faculty advisors and voted on by their GRASP peers.

Ahmad Amine (PhD, ESE)

SIT-LMPC: Safe Information-Theoretic Learning Model Predictive Control for Iterative Tasks

Iterative tasks are common in robotics applications ranging from pick-and-place tasks to autonomous racing. Finding optimal trajectories for these tasks has been thoroughly explored in literature through techniques like trajectory planning, optimal control, and reinforcement learning. Learning Model Predictive Control (LMPC) solves for the optimal trajectory that maximizes performance for a given task by learning the value function from trajectories executed in previous iterations. Unlike reinforcement learning, LMPC provides desirable theoretical properties for safety and convergence, which is crucial for safety-critical applications. We present an approach for extending the LMPC problem to stochastic nonlinear systems using Model Predictive Path Integral (MPPI). Previous methods utilizing optimization-based techniques or Cross-Entropy-Method sampling to solve the LMPC problem fall short when dealing with high-dimensional, nonlinear dynamics. Our MPPI-based framework optimizes the control policy from an information-theoretic perspective and overcomes these limitations by providing a systematic way of handling constraints without sacrificing sample spread. This is achieved by incorporating penalty functions and optimizing over multipliers to balance out constraint satisfaction with controller performance. The proposed algorithm can be parallelized allowing for real-time control on limited compute hardware. We validate our approach through simulations and real-world experiments, demonstrating significant improvements in constraint satisfaction and final trajectory performance.

Erica Santos (MSE, ROBO)

Temporarily, Robots Unite to Surmount Sandy Entrapments, then Separate (TRUSSES)

The TRUSSES project is developing methods for teams of robots to jointly overcome environmental hazards on the Moon by attaching to each other to form larger and more stable, maneuverable structures. The robots will use their interactions with the ground to form a map of safe and risky terrain, attach to each other as support when the ground traversal risk is high, move in a coordinated fashion once joined, and, once the maneuver has been successfully completed, separate to continue their original individual missions.

Darshan Thaker (PhD, CIS)

Frequency Guided Posterior Sampling for Diffusion-based Image Restoration

Image restoration aims to recover high-quality images from degraded observations. When the degradation process is known, the recovery problem can be formulated as an inverse problem, and in a Bayesian context, the goal is to sample a clean reconstruction given the degraded observation. Recently, modern pretrained diffusion models have been used for image restoration by modifying their sampling procedure to account for the degradation process. However, these methods often rely on certain approximations that can lead to significant errors and compromised sample quality. We present a simple modification to existing diffusion-based restoration methods that helps mitigate these errors based on a frequency analysis of the degraded observations. This results in significant improvement on challenging image restoration tasks such as motion deblurring and image dehazing.