The course was unknown by design. The robot could not rely on a map; it had to discover the route while it was moving.
The problem
BotBoost was a university project built around autonomous navigation in an unknown obstacle course. As lead developer and project manager, I coordinated the team's work from algorithm selection through simulation tests and the final competition.
We used Python 3.8 and ROS to connect the control program with TurtleBot hardware. LIDAR and GPS data supplied the inputs for real-time navigation, while GitLab and Ubuntu gave us a shared development and test environment. Keeping the sensor integration, control logic, and simulation work separate made it possible to compare approaches without rebuilding the whole system each time.
The build
A* was one of several control algorithms we implemented and benchmarked. The final robot used a Bug v2 wall-following strategy, with front and side sensor thresholds tuned for the course. We evaluated the options in simulated trials before using them in the competition, focusing on how well each approach handled obstacles, route choice, and response time. This was not a case of picking an algorithm because it was well known. We used the benchmark results to decide what belonged in the final control program.
The implementation details were where the project became real. The front sensor threshold was 1.2 metres and the left sensor threshold 1.5 metres. The robot could only declare itself back on the line after travelling at least 0.4 metres, which prevented it from oscillating between obstacle detection and line joining. We also limited acceleration because a sudden move above 0.2 m/s made the robot drift roughly 30 degrees to the right.
The simulation phase gave us a reproducible place to compare controllers before we committed time on the physical robot. It also exposed assumptions about sensor range that would have been expensive to discover during a run. We still had to tune the final thresholds on the TurtleBot, but the simulations narrowed the search.
The result
We built a ROS-based TurtleBot navigation stack and benchmarked control algorithms for an unknown obstacle course. The robot could accept published goals, navigate around obstacles, and return to its route. The project was an early lesson in systems thinking: pathfinding on its own is not enough. Sensor quality, real-time control, and how the team tests changes determine whether the robot actually reaches its destination.
What I took from it
Leading the project taught me to make roles and test criteria explicit when people bring different levels of robotics experience. If I revisited it today, I would formalise benchmark scenarios and capture more telemetry from each run. The core experience remains valuable because it connected AI concepts, robotics hardware, and collaborative delivery under a fixed deadline.