RESPONSIBLE INTELLIGENCE · 02
How we approach physical intelligence.
A robotics reasoning system is only useful when people can understand its state, set its limits, and intervene. It also has to be suitable for the machine, task, and environment where it operates.
Our operating principles
People must be able to supervise, interrupt, and define the operating limits of deployed systems.
Machine decisions should remain within explicit hardware, environment, force, and task constraints.
Predictions should be checked against sensed outcomes so uncertainty and physical change remain visible.
Systems should identify degraded sensing, unfamiliar conditions, and confidence limits before escalating action.
Validation before autonomy
Ontos evaluates a system on the machine, task, and environment where it may operate. Testing can cover sensor quality, latency, force and motion limits, repeatability, uncertainty, recovery behaviour, and the ways a person can intervene. A demonstration is not the same as production validation.
Privacy by system design
Physical-world sensing can reveal information about people and places. Ontos aims to minimise unnecessary collection, keep processing close to the device where practical, restrict access, define retention, and make sensing states understandable to operators. Camera use, if present in a future product, should be disclosed and governed by the same principles.
Security and resilience
Connected hardware and reasoning software should be designed for authenticated access, protected updates, monitored failure states, and recovery. Potential website security issues can be reported through the Security page.
Claims and development status
Ontos distinguishes product direction, prototypes, demonstrations, and validated capabilities. Performance claims should identify their test conditions and should not imply that an experimental system is ready for unrestricted deployment.
Questions
Questions about responsible development can be sent to hello@ontos.ws.