Safety-aware Expansion for Neural Network Repair

iROS'23

Inspired by the popular cascade-correlation algorithm, this project introduces a safety-aware growing neural network method aimed at expanding neural networks with new hidden units dedicated exclusively to responding to faulty samples.

Safety-aware Neural Network Repair for Robotic Systems with Predictive Models

IROS'2024

This project presents a two-step supervised learning approach for safety-aware robot learning, using predictive models to enforce explicit and implicit contraints on the policy.

Safe Robot Learning in Assistive Devices through Neural Network Repair

CoRL'22 - NeurIPS'22

This project presents NNRepLayer: the first repair technique applied to a real physical system.

Joint Communication and Motion Planning for Cobots

ICRA'21

This project unifies communication and motion planning to enrich human-robot interaction, to prevent potential deadlocks and freezing situations, and to maintain the planning efficiency in terms of raveled distances.

Risk-bounded Motion Planning and Control in Confined Dynamic Environment

iROS'21 - L-CSS'20

This project addresses the "Freezing Robot" problem in dynamic confined environments. We introduced a method to bound the probability of collision to a given value in finite-time.

Kinematic-based Trajectory Planning Framework for Autonomous Ground Vehicles

IEEE-JAS'19 - IV'18

This project presents an optimization framework for the trajectory planning problem of autonomous vehicles for a bounded time horizon.