A portion of our engineering hours every quarter is reserved for fundamental research — the work that won't ship next year, but might rewrite what's possible the year after.
Cutting the gap between virtual and physical training cycles. We're particularly interested in reducing the number of physical rollouts required to fine-tune a sim-trained policy on a new platform — currently published results need thousands; we're aiming for tens.
Vision tells you where something is. Touch tells you what it is, and how it's reacting to you. We're building dense sensorized skins and the inference pipelines to use them — for manipulation that doesn't fail when the lights go out.
Centralized planners don't scale. We're studying gossip protocols, emergent role assignment, and partial-observation consensus for fleets that can lose 20% of their members and still complete the mission.
We co-author with universities, host visiting researchers, and run a small fellowship program for ambitious PhDs and recent post-docs.