Bayesian localisation and path-following control for a Mecanum-wheeled robot, built for MECHENG 313 (Design of Real-Time Software) — the course I was awarded Top in Course for.

the problem

The robot starts from an unknown pose inside a walled arena and has to figure out where it is, then execute a full rectangular coverage path — all in real time, on embedded hardware, fusing noisy sensor data.

approach

• Localisation. Bayesian localisation fusing IMU and IR sensor data lets the robot self-orient from an unknown starting pose.
• Control. A closed-loop PID controller with encoder feedback drives position control to zero steady-state error.
• Coverage. Path-following control executes a complete rectangular coverage sweep of the arena.

takeaways

Tuning the estimator and controller together — on real, noisy hardware with hard real-time constraints — was the core challenge and the most valuable part. It’s where sensor fusion, control theory, and real-time software all had to actually agree.