Attention hourglass

Robotics attention has a weird hourglass shape: two big ends and a thin middle. One end is the flashy demos, the industry's business card for outsiders. The other is data collection, the entry point for people coming in from other industries. The middle is everything needed to turn field data into a deployed robot (training and architectures, inference and control, generalization, evals, deployment infrastructure, hardware) and it holds many of the hardest unsolved problems while getting surprisingly little attention. We have collectively over-indexed on data as the thing holding back the field. Data is a bottleneck, but not the only one. That middle is where I'm looking for my way back in, particularly new training techniques and architectures and the infrastructure for turning physical data into reliable behavior.