BlindBike

Most assistive cycling hardware assumes you will buy a special bike. This is an attachment: it clamps to the bike someone already owns, senses what is in front of them, and says so through the grips.

The board is dual-MCU. An STM32G474 runs the sensing and haptics loop, and an ESP32-S3 handles the camera, audio and connectivity. Splitting them is the whole point — a hung network stack cannot stall the thing that is supposed to warn you about a bollard. LiDAR, radar, IMU and a vision model are fused to flag a hazard in time to matter, and the schematic is ERC-clean in KiCad.

I validated the problem before I drew the board, through user research at a school for the blind. That is where the decision to put feedback in the grips came from rather than audio alone: audio competes with traffic, which is the one thing a rider cannot afford to stop hearing.

There is also a ROS 2 and Gazebo simulation whose hazard logic mirrors the firmware, so the decision layer can be exercised without putting a rider on a road.

Archived rather than killed. The design is finished and what stopped it was my time, not the evidence — which is a different thing from what happened to RD-02, and worth saying plainly.