QERRA-v2 Classical adds deterministic ethical safety layer for Behavior Trees
No LLM, no ROS2 needed — deterministic ethical gatekeeping for robots goes live.
QERRA-v2 Classical is an open-source, deterministic ethical evaluation layer designed for robotics Behavior Trees. Built entirely without large language models, it uses the SEMEV-12 framework — 12 explainable ethical vectors covering coercion, autonomy violation, harm intent, and more. A companion physical-safety layer (QERRA-HSR) detects acute distress, isolation, and environmental hazards, running with higher priority: if a critical safety issue is flagged, the ethical evaluation is suspended entirely. The system outputs a full reasoning trace for every decision, including which vectors fired and similarity scores, making it auditable by design.
The latest release adds a standalone PyTrees Condition node, enabling integration without any ROS2 installation — just pip install py_trees and requests, set an API key, and run a test script. The test covers two scenarios: a safe action that executes normally, and a high-risk autonomy violation that the tree correctly blocks and routes to a human review fallback. All interactions go through a public API (TEST key provided). The developer, a solo contributor, is actively seeking feedback from the Behavior Tree and robot safety community on interface design and missing features.
- Two-layer architecture: QERRA-HSR for physical safety (collapse, isolation, hazards) runs first and can suspend SEMEV-12 ethical evaluation if critical
- SEMEV-12 evaluates 12 explainable vectors (coercion, autonomy violation, harm intent) with deterministic reasoning — no LLM in the decision path
- New standalone PyTrees Condition node works without ROS2 setup: pip install py_trees requests and run test_bt_tick.py against the live API
Why It Matters
Enables deterministic, auditable ethical gatekeeping for autonomous robots — critical for safety-critical deployments without black-box AI.