Publications & Research

A unified framework coupling discrete-time CBF and CLF-guided reverse diffusion with iterative-projection MPC for safe robot trajectory generation under dynamic constraints, validated on a physical 7-DOF Franka manipulator.

A diffusion-based CRL framework that adapts mask-conditional image inpainting on trajectory heatmaps to automatically generate achievable, progressively challenging curriculum goals.

An interpretable fuzzy-logic reward shaping method that adaptively shapes velocity-distance rewards using human-like linguistic rules, demonstrated on autonomous drone racing in IsaacLab.

A practical expert-distillation pipeline transferring RL and optimal control demonstrations into compact Vision-Language-Action (VLA) models for language-conditioned aerial drone navigation.

Generative CRL framework leveraging conditional diffusion models conditioned on Q-value feasibility and AIM reward proximity to synthesize optimal curriculum goals.

Prioritizes replay buffers using contact energy and tactile forces measured at the robot gripper, significantly expediting robotic manipulation policy learning.

A coordinated motion planning and hierarchical optimization control framework for high-degree-of-freedom redundant mobile manipulators in constrained spaces.

Proposes an environment-shifting curriculum learning paradigm that shifts physical parameters progressively to tackle sparse reward manipulation challenges.

Combines multi-objective evolutionary search (NSGA-II) with hindsight experience replay to optimize value prediction, goal-proximity potential, and Wasserstein distance under obstacle constraints.

Evolves interpretable binary decision trees via Grammatical Evolution and continuously refines waypoints using critic Q-value feedback for sparse-reward manipulation.

Co-evolves morphological and behavioral traits of miniaturized swarm sensory agents exploring uncharted environments.