dum-e is my test bench for robots that can do a little more on their own.
Motor control, sensing, mapping and autonomous behaviors. Built on the bench, tried in simulation, and documented with the measurements that make the result worth sharing.
Read the latest experiment ↓
The aim is a robot that can explore, understand enough of its surroundings to act, and make useful decisions onboard. ROS 2 connects the pieces; simulation provides a place to test them before the hardware does.
This notebook follows that work as it happens: controller experiments, integration details, tools and the occasional fix needed along the way. Each study keeps its code, method and evidence close together.
Could a tiny neural network steer a robot using weights written directly by an AI? Two rooms, one doorway and a lidar make the question concrete. Recorded runs and a comparison of architectures show what it can do.
Read the experiment ↗