High-Traversability and Precise Navigation for Mobile Robots in Constrained Environments

This article presents an integrated sensing-planning navigation system for mobile robots used in industrial inspection. The system can be directly integrated with real robots. It enables the robot to pass through tight spaces, circumvent unexpected obstacles, and accurately reach the goal pose. This improves the robot’s adaptability to inspection tasks. For obstacle sensing, the system includes an obstacle parameterizer. Static obstacles in global grid map occupancy clusters are convexified. To improve the representation accuracy of the tight space, continuous and angled obstacles are automatically segmented into multiple convex polygons. This avoids inappropriate convex shapes. Unexpected obstacles in local grid maps are first represented by minimum bounding circles (MBCs) to simplify data association. They are then converted into inscribed convex polygons to ensure consistency with static obstacle descriptions. For motion planning, the system utilizes a two-stage planner operating on obstacle convex sets. When the robot is distant from the goal, the nonlinear model predictive control (NMPC) is constrained by interconvex set discrete-time control barrier functions (DCBFs-Convex), termed NMPC-DCBF-Convex, enabling effective navigation in tight spaces and around unexpected obstacles. As the robot nears the goal pose, DCBF-Convex conditions prune the sampling space of dynamic window approach (DWA), termed DWA-DCBF-Convex. It ensures high-frequency control and enables safe and precise arrival at the goal in constrained environments. Experiments in constrained scenarios validate the system’s real-world effectiveness. An inspection robot weighing over 70 kg completes challenging navigation tasks at a maximum velocity of 0.5 m/s, achieving an empirical terminal positional error of ±0.017 m.