Intent-Aware Dynamic Window Planning via Bayesian Goal Inference and Control Barrier Function Safety Filtering for Shared-Control Telepresence Robots

Shared-control telepresence navigation typically blends operator inputs with local planners like the Dynamic Window Approach (DWA), yet most schemes are not intent-aware and lack formal safety guarantees. The study proposed Intent-Aware Dynamic Window Planning via Bayesian Goal Inference and Control Barrier Function (CBF) Safety Filtering. Operator intent is modeled as a latent goal over a discrete hypothesis set and inferred online via a Bayesian filter from joystick velocities and workspace geometry. The implied intent shapes the DWA objective through goal-aligned cost terms while a zeroing CBF imposes forward-invariant safety sets that constrain admissible velocities, decoupling performance and safety. We implement a Python simulation of unicycle dynamics with realistic sensing, moving obstacles, and injected network latency (0–200 ms). Benchmarks span 1,200 randomized indoor maps. Baselines include teleoperation, classical DWA, and goal-biased DWA without CBF.