Ideas

Compliant joints using SEAs could dramatically improve energy efficiency and terrain adaptation in legged robots. The key challenge is spring sizing — too stiff and you lose compliance, too soft and control bandwidth suffers. Worth exploring with a simple 2-DOF leg testbed before committing to a full platform.

Intel RealSense or OAK-D cameras are cheap enough now that integrating one into a ground robot for live terrain maps seems viable on a real budget. The open question is compute — running depth fusion on a Raspberry Pi 5 versus offloading to an edge compute module. Need to benchmark both before deciding.

Using a PCB as a primary load-bearing structural element instead of an afterthought mounted to a separate chassis. FR4 has reasonable stiffness. The interesting design challenge is co-optimizing board layout and mechanical constraints from the start, rather than routing around predetermined geometry.

A mesh of motor controllers communicating over LoRa or 2.4GHz with a central coordinator. Eliminates CAN bus wiring complexity in large multi-joint systems. Latency and reliability are the obvious concerns — for fast control loops this probably doesn't work, but for slower manipulators it might be worth the weight and wiring savings.

Most IMU fusion algorithms assume a well-calibrated accelerometer and gyroscope. What if the filter was designed to be robust to calibration drift from the start? Useful for cheap MEMS IMUs deployed in the field where recalibration isn't feasible. Probably involves adaptive covariance estimation in the EKF.