Towards Trustworthy Autonomous Robots: An Explainable AI-Based Decision Framework

TL;DR

The paper introduces TRACE, a framework that enables full auditability of autonomous robot decisions by linking actions to sensor data through structured causal chains. It organizes decision-making into four layers: perception, belief reasoning, action synthesis, and execution verification, ensuring transparency without restricting the use of deep learning models. Evaluation on warehouse navigation shows high performance across traceability, reconstructability, and continuity metrics.

Why It Matters

Trustworthy autonomous systems are essential in safety-critical environments like warehouses and healthcare. TRACE enables regulators and operators to verify decisions after incidents, meeting legal standards such as the EU AI Act. This enhances accountability and supports the deployment of AI in high-risk applications.

Key Findings

• TRACE achieves 98.6% evidence traceability by maintaining sensor-to-decision links through causal chains.
• Decision reconstructability reaches 99.0% due to structured post-hoc analysis enabled by documented counterfactuals and belief states.
• The framework maintains temporal continuity at 98.1%, ensuring complete audit trails across decision cycles.