Back to the scenes
Robsys Robotic Systems, Mar 2024 to Mar 2025
Autonomous navigation for a cleaning robot
Built SLAM, pathfinding and obstacle avoidance from scratch, because off-the-shelf SLAM frameworks did not run on the robot's hardware.
Problem
A cleaning robot had to map rooms, plan routes and avoid obstacles on its own. The off-the-shelf SLAM frameworks did not run on its hardware.
Approach
- Navigation from scratch. Custom SLAM, pathfinding and obstacle avoidance.
- Sensor fusion on NVIDIA Jetson. Stereo cameras, gyroscopes and ultrasonic sensors in one pipeline, custom deep learning models for environmental perception, and low-level firmware for sensor synchronization and motor control.
- Talking to the robot. A UART protocol, co-designed with an embedded engineer, for two-way data exchange between the robot’s motherboard and its subsystems.
- Remote operation. Real-time telemetry and camera streaming into a mobile app for monitoring, control and safety operations.
Result
A navigation system that runs on the robot’s own hardware, with live telemetry and camera streaming to a mobile app.
What I would do next
Benchmark the custom SLAM against a reference trajectory on recorded runs, to see where map drift comes from.