Simulation · Project case study

Autonomous navigation

A Pioneer 3-AT navigation system combining A* planning, DWA obstacle avoidance and position-based recovery in Webots.

DisciplineSimulation
StatusCompleted
Year2026

Project overview

PythonWebots R2025aPioneer 3-ATA*DWALiDAROccupancy gridRecovery control
6 / 6Reported simulation trials reached the goal · 44.7 s mean
Webots simulation arena — 37 cylindrical obstacles

I developed a Python navigation system for a simulated Pioneer 3-AT in Webots R2025a. The robot navigates a 60 × 60 m arena containing 37 static cylindrical obstacles, using a SICK LMS 291 lidar, GPS and compass. A separate supervisor controller records position and completion metrics.

Watch presentation

The engineering

The challenge

Reach a fixed goal through a cluttered arena while avoiding obstacles and recovering when the reactive local planner stops making progress.

My contribution

Integrated planning, lidar mapping, wheel control and recovery; iterated the stalled-robot response and evaluated six simulation trials with a separate supervisor.

Approach & implementation

A* supplies global waypoints from a lidar-updated binary grid. DWA selects local motion commands, while displacement-based stuck detection can override it with a reverse-and-turn recovery. A separate Webots supervisor records outcomes and timing.

Results & lessons

What the project achieved

All six reported trials reached the 1 m goal acceptance radius before the 120-second timeout. The six tabulated times average 44.65 seconds, rounded to 44.7 seconds. Recovery activated seven times across trials 3 and 6; both trials ultimately reached the goal. This is a small evaluation in one fixed simulated arena, not a general reliability guarantee.

Limitations & next steps

The simulation uses ground-truth GPS/compass pose, noise-free lidar and static obstacles. Binary occupied cells can become stale, and eight-connected A* paths contain sharp corners. Recovery addresses observed stalls but does not anticipate dead ends. The report’s suggested 67% success without recovery is a hypothetical comparison, not a measured ablation. Next steps include larger randomised trials, sensor noise, probabilistic mapping, SLAM, smoother planning and dynamic-obstacle evaluation.

Explore the engineering

Tap cells to add obstacles, then find a path.

Simplified interactive concept demo: A* on a four-neighbour grid. The reported Webots project uses eight neighbours, including diagonals. S = start, G = goal. This explains global planning; it is not a replay of the Webots project or its DWA controller.

Media & resources

Let’s talk about this project.

I’m happy to walk through the design decisions, challenges and what I would improve next.

Contact Peter