96%
Successful interventions
VL53L7CX
Multizone ToF sensors
12.2%
Average CPU usage
The RELbot
The RELbot is an educational robotics platform that students use to develop control, AI, and vision-based navigation algorithms. The beauty of this system is that it operates on Xenomai, a real-time OS. This allows students to experiment with real-time challenges and truly optimise their algorithms.
The standard assignment for this system is to create a computer vision algorithm that tracks a green ball and follows it. While these environments are ideal for teaching robotics, they also introduce the risk of collisions and accidental drop-offs. A single programming error can cause the robot to drive into objects or fall from an elevated platform, resulting in damaged hardware.
This project focused on designing and implementing a dedicated safety layer that can intervene independently of the user's software. The goal was to create a predictable and minimally intrusive system that protects the robot without restricting experimentation. The idea is to allow students to make mistakes in a safe environment.
Figure 1 — Front view of the RELbot.
Multi-Zone Time-of-Flight Sensing
Research identified multi-zone Time-of-Flight sensors as a promising solution for obstacle and floor-edge detection. Two VL53L7CX sensors were integrated onto the RELbot to create a wide forward field of view.
The two multi-zone sensors provide an 8×8 distance grid, allowing the system to distinguish between floor surfaces, drop-offs, and nearby obstacles.
To maximise floor-edge visibility, custom mounts were designed that position the sensors on the roof of the RELbot at a downward angle. A calibration procedure was developed to compensate for mounting tolerances and generate a reference map of the surrounding environment. This allows the RELbot to differentiate objects that have collision risks and objects it will drive past.
Rather than using a fixed activation distance, the safety layer employs a velocity-dependent threshold model. As the RELbot moves faster, its braking distance increases, causing the safety zone to expand automatically. This dynamic threshold ensures that the robot maintains the minimal stopping distance at all operating speeds.
Figure 2 — Sensor Mounts for the RELbot.
Real-Time Safety Architecture
The safety layer was designed to integrate seamlessly with the existing XRF2 framework, which combines ROS2 running on Ubuntu with the firm real-time Xenomai environment.
Sensor acquisition, calibration, and measurement analysis are performed within ROS2, while the final safety decision is executed in Xenomai. This separation allows computationally intensive sensor processing to run independently from the real-time control loop, while ensuring that emergency braking remains deterministic and responsive.
When an obstacle or floor-edge enters the safety zone, the safety layer temporarily overrides the user's controller and applies an emergency motor lock. This binary response was chosen to maximise predictability and minimise complexity. Rather than attempting to predict user intentions or modify trajectories, the system only intervenes when a collision or drop-off becomes imminent.
To provide clear feedback to users, a red LED indicator activates whenever the safety layer takes control, making the intervention immediately visible during operation.
Figure 3 — Safety layer preventing a collision.
A safe enviroment
The safety layer successfully created a safe platform for students to make mistakes while learning robotics. By combining multi-zone ToF sensing, dynamic safety thresholds, and real-time emergency braking, the system provides reliable protection against both collisions and drop-offs while remaining largely invisible during normal operation.
Testing showed that the safety feature successfully intervened in 96% of evaluated scenarios while maintaining an average CPU usage of only 12.2%. The resulting system exceeds the project requirements and demonstrates that advanced safety functionality can be added to educational robotics platforms without compromising usability or performance.
Safety-layer design for the RELbot
Veldwachter2025asp.pdf