Device integration and simulation on a real rover.
Big Stu is an operator-controlled, four-wheel rover. The project combines ROS 2 motor control, sensor recording and a simulation model, with recorded runs used to compare the model’s behaviour with the physical rover.
Connecting motor controllers to ROS 2.
Four hub motors are driven by two motor controllers over Modbus RTU. We wrote a Linux driver from the controller’s register manual and connected it to a custom ROS 2 node.
The node translates velocity commands into wheel setpoints and reports motor speed, current and faults. The implementation includes command-timeout handling, acceleration ramps and fault decoding.
This is the type of interface work a device integration involves: understanding the protocol, handling device state and making behaviour observable during tests. Our ros2_control interface has also been used to drive the robot, with velocity commands, wheel feedback and fault reporting. Its project page describes the implementation and tests.
Recording what the machine sees and reports.
Camera, LiDAR, inertial and drive data let us revisit a run and compare sensor readings. This clip shows the LiDAR stream during an operator-driven pass, with the rover centred in the view and the surrounding rows moving past it.
Checking model assumptions against measurements.
The rover and its test environment are also modelled in Isaac Sim. Comparing the model with field recordings exposed differences that affect how its results should be used.
Environment geometry
The original simulated rows were 3.05 m apart. A survey of six recordings across four sessions gave a median of 2.30 m. The field model was rebuilt using the measured spacing.
Motion response
The model has a documented steering-response mismatch. Its simulated motion cannot be used as proof of physical tracking performance.
Scope of an assessment
A model assessment identifies what was compared, the test conditions and the differences found. Whether a model is suitable depends on its intended use and the agreed criteria.
Updated field scene with detailed blueberry models. Scripted motion illustrates the scene; the measurements discussed here come from the earlier comparison runs.
Need a device interface or a model assessment?
Bring the device documentation, the problem you need to resolve and any existing recordings. We can discuss a bounded scope and the checks the work would need.