Driving without considering the preferred separation distance from surrounding vehicles may cause discomfort for users. To address this limitation, this paper proposes a human-preference-aware driving planner that incorporates users’ desired clearance from surrounding vehicles as constraints in an optimal control problem. Simulations with different user responses show that the planner better aligns with individual preferences than preference-agnostic baselines.

Crowded and dynamic environments require fast replanning; however, single-trajectory optimization lacks alternative paths, while parallel optimization often relies on predefined initial guidance. In this work, we identify multimodal optimal trajectory distributions without initial guidance by projecting sampled trajectories onto safe constraint sets, clustering them, and optimizing each cluster using MPPI. Simulation and hardware experiments demonstrate higher success rates and safe drone navigation in crowded 3D environments.

Denoising-based planning can generate safe and optimized trajectories but often requires slow iterative refinement. This paper proposes a normalizing-flow-based path-planning method that refines trajectories step by step along the planning horizon, enabling adaptive trajectory-length adjustment under computational time constraints. The generated trajectory can also initialize MPPI optimization, reducing required iterations and improving efficiency. The method is validated through simulation comparisons with diffusion-like planners and hardware experiments.

Multi-robot systems improve efficiency through cooperation, but collision avoidance with static and dynamic obstacles remains a major challenge. This survey reviews recent collision avoidance methods, including classical centralized and decentralized approaches as well as learning-based methods.

To improve sampling efficiency in sampling based trajectory optimization, we focus on increasing the ratio of feasible trajectory samples among all sampled trajectories. To this end, we propose LRFS MPPI, which combines the Lateral Recursive Feasible Set (LRFS), a recursively computed feasible steering input set that guarantees the vehicle remains within road boundaries over the prediction horizon, with truncated normal sampling. We demonstrate through simulation and real world experiments that the proposed method generates high quality collision free trajectories more efficiently and reliably.