Keyvan Majd
Research Projects
NNRepLayer: Neural Network Repair for Safe Robot Learning

NNRepLayer: Neural Network Repair for Safe Robot Learning
CoRL'22

Framework for repairing neural network control policies to satisfy formal safety constraints using Mixed Integer Quadratic Programming. Applied to robotic prosthesis control. Best Paper Award at NeurIPS'22 Robot Learning Workshop.
Safe Navigation with Control Barrier Functions

Safe Navigation with Control Barrier Functions
IROS'21

Extended time-based RRTs with Control Barrier Functions for safe motion planning in dynamic environments with pedestrians. Demonstrated on Toyota HSR in narrow corridors.
Open X-Embodiment: Large-Scale Robot Learning Dataset

Open X-Embodiment: Large-Scale Robot Learning Dataset
CoRL'23

Contributed to the Open X-Embodiment dataset — a large-scale multi-robot learning dataset supporting policy learning across diverse embodied platforms, enabling cross-embodiment generalization.