New AI framework combines trajectory optimization with deep reinforcement learning to create safer, more comfortable navigation assistance for visually impaired users.

MHHTOF framework balances safety and comfort objectives using Momentum-Constrained Trajectory Optimization to reduce sudden velocity and acceleration changes

Residual-enhanced deep reinforcement learning module improves trajectory refinement with better temporal modeling and generalization across scenarios

Dual-stage cost modeling mechanism uses both Frenet space consistency checks and Cartesian space reward-driven weights to integrate user preferences into navigation planning