01
See the course
A five-channel IR array measures line position, while wheel encoders and the BNO055 IMU provide motion and heading feedback.
Mechatronics / Autonomous systemsJanuary - March 2026
Romi autonomous navigation robot
A two-wheel robot that follows lines, estimates its position from encoders and an IMU, detects a wall without a bump switch, and sequences its own route through a mixed-mode course.
05
scheduled tasks
20 ms
observer period
10
navigation stages
Final result
One course, multiple control modes.
The final run combines ordinary line tracking with waypoint checks, precise turns, distance moves, and acceleration-based wall contact. The controller coordinates the transitions without a human selecting each mode.
System design
Feedback from floor to firmware.
Controller
STM32 Nucleo L476RG
Firmware
MicroPython
Feedback
BNO055 IMU + dual encoders
Line sensing
5-channel digital IR array
Control
PI/PID + state estimation
Runtime
Cooperative task scheduler
Control architecture
Small tasks, shared state, one coordinated machine.
02
Estimate motion
An observer running every 20 ms fuses motor effort, encoder position, IMU heading, and yaw rate into a live pose estimate.
03
Choose the behavior
A finite-state controller switches between line following, heading control, distance control, and impact-based wall detection.
04
Drive the motors
Independent 50 ms motor tasks apply the requested effort and report encoder data through shared variables and queues.
Test progression
Prove the loop, then add the course.
Stage 01 / PI tuning
Continuous line-following test
A circular course exposed oscillation quickly and made it practical to tune base speed, proportional gain, and integral gain before attempting the full route.
Stage 02 / autonomous route
Mixed-mode final track
The course demanded long line segments, curves, a garage maneuver, wall contact, heading changes, and an unlined finish. A ten-stage hub sequenced each behavior.
Engineering lessons
The useful parts were not all victories.
Battery voltage changed the plant
The same motor command behaved differently as the battery discharged. A future revision would measure supply voltage and compensate the effort automatically.
Digital sensing made the line coarse
The digital array was quick to integrate, but its five widely spaced sensors produced abrupt corrections. A denser analog array would give the controller a smoother error signal.
The IMU doubled as a bump sensor
Instead of adding a mechanical switch, the robot detected a wall through acceleration. It worked, but hard acceleration could mimic an impact and required careful threshold tuning.
Finite-state design
The behavior behind the motion.
Team credit
Built together.
This project was created by Shafiq Amat and Max Soury. The firmware, testing, documentation, wiring, diagrams, and physical robot were the result of a shared team effort.