Every robot action can be traced back to the sensor evidence that caused it.

Autonomous robots that rely on deep learning have a major flaw: when accidents happen, human investigators cannot tell why the machine picked a specific action. To solve this, researchers built TRACE, a new decision framework. It works by routing every robotic choice through four clear, traceable layers. First, it identifies objects using sensor proof. Next, it tracks states using cause-and-effect models. Then, it plans actions while noting alternate choices, and finally checks that the robot follows the rules. In tests on simulated warehouse robots over 500 decision cycles, TRACE connected actions to sensor evidence 98.6% of the time. It also hit 99.0% for audit trail completeness and 98.1% for the ability to rebuild past decisions, providing the accountability safety-critical robots need.

Why it matters

When an autonomous robot causes an accident, investigators can now reconstruct why it acted that way. This improves safety in places like warehouses and hospitals, and helps companies comply with laws like the EU AI Act that require transparency for high-risk systems.

Questions answered