OhhO Mind
OhhO Autonomy lets a robot follow a mission; OhhO Mind lets it decide on one. Mind is the deliberative layer that runs continuously on top of your stack — perceive → reason → verify → act → monitor → reflect → remember — turning a capable robot into an agent you give goals to, not scripts.
Give your robot a mind of its own.
Foundation models gave robots brilliant reflexes — a Vision-Language-Action model turns an image and an instruction into an action. But a reflex isn't a mind. A robot still can't choose a goal, remember what it saw, notice that it failed, or get better over time. OhhO Mind is the layer that closes that gap.
Mind runs a continuous decision loop on top of the robot's existing fast control stack. It fuses everything the robot knows into one world-state snapshot, reasons over it to pick the next action, verifies that action against the robot's real hardware limits, executes it through OhhO Autonomy and OhhO Serve, watches the outcome, and reflects on whether the goal was met — then remembers what it learned and does it again. You hand it an objective in plain language; it loops until the job is done, then idles.
Mind on this site is a prototype console plus the agent_engine and learning_engine packages. Cloud, on-device, and NPU routing is roadmap. Over-the-air delivery to a fleet is roadmap. What you can run today is the unit tests and the simulated console.
- The agent loop. A continuous perceive → reason → verify → act → monitor → reflect → remember cycle. Give it a goal; it runs until done, recovers from failure, and idles when there's nothing to do.
- Grounded world state. Mind fuses pose, arm state, detected objects, mission status and memory into one snapshot — so it reasons about the world the robot is actually in, not a guessed one.
- A safety gate it can't skip. Every low-level action is checked against the robot's real limits — velocity, joint deltas, reach — before it reaches a motor. Unsafe plans are rejected, not clipped.
- Hybrid reasoning. Cloud-class reasoning when online, an on-device model and NPU when offline. The robot degrades gracefully instead of going dark when the network does.
- Memory that compounds. Mind remembers objects, places and the outcomes of past attempts, and feeds them into every decision — so the robot builds a model of its own environment.
- Learns from experience. Each attempt is judged, labelled and stored, then fed to OhhO Train to improve the policies — closing the loop from operation back to capability.
Related: OhhO Autonomy, OhhO Serve, OhhO Train