On this page

Keep the robot’s requirements, changes and evidence connected.

Physical AI systems combine perception, control software and hardware in a changing physical environment. Their engineering record needs to preserve those dependencies, along with the configuration and conditions behind each result.

Review the whole sensor change.

In a fictional warehouse robot, a camera replacement can affect image timing, a perception input interface, controller assumptions and the conditions represented by a stopping test. Start with those connections rather than treating the replacement as a parts-list update.

Fictional review example; not executed product or customer evidence.
Engineering record	What the owner checks
Operating requirement	The intended speed, visibility and surface conditions remain explicit.
Sensor / perception interface	Timing, timestamps, calibration and input assumptions remain compatible.
Control and hardware	The controller assumptions and physical response still support the requirement.
Verification evidence	The result identifies the sensor, software/model version and conditions actually tested.

Arc connects these records, exposes related context and supports proposed changes on a branch for review. Engineers decide what needs analysis, testing or acceptance. Arc does not train robot models, control robots at runtime or certify a system.

Explore four engineering workflows.

Connect requirements across the robotTranslate operating conditions into requirements, interfaces and verification responsibilities.Review a model substitution in contextReview M1-to-M2 dependencies, requirements and evidence against the same proposal revision.Keep results tied to configurationsRecord the hardware, software, model and conditions a test result actually supports.Map CI evidence for reviewUse a conceptual CI evidence mapping to connect reports, revisions and configurations, with recovery and engineering review kept explicit.

Connect your tooling with a clear boundary.

Use the REST API for custom test and delivery integrations, or MCP for compatible AI assistants working within programme permissions. Keep test artefacts in approved storage and review the configuration record before accepting a linked result.

Standard Arc is hosted on AWS in London. External AI processing may occur elsewhere; no-model-training terms do not establish zero retention. On-premises, isolated-network and local-inference options require supported arrangements and separate terms. Review security and deployment →

Bring one engineering change.

Evaluate Arc with a requirement, its affected interfaces and a representative evidence record. Agree scope and data handling before adding sensitive programme information. Luc will follow up to arrange setup.

Request a trial →Explore Product →

Next reading

Requirements Management for Robotics and Physical AI: From Task to Acceptance: Complete the twelve-field handoff for a measurable replay timing requirement and its owned gaps.
Robotics Change Impact Analysis: Which Requirements and Evidence Need Review?: Review the M1-to-M2 substitution through seven dependency domains and open evidence actions.
Robotics Verification Evidence: Linking Requirements to the Tested Configuration: Judge which tested configuration and conditions a result supports before proposing reuse.
Physical AI Test Evidence in CI: Connecting Results to Requirements and Reviews: Prepare a conceptual CI evidence proposal and reconcile uncertain submission state under a reviewed contract.