中文：危害反射宪章式架构：面向具身机器人的人造神经身体框架
Robots need an artificial nervous body, not only an AI brain.
机器人不仅需要 AI 大脑，也需要人造神经身体。
Status: Public design framework (v0.2.1). HRCA is not a finished robot controller, certified safety system, product, benchmark, or proof of safety. It is a conservative architecture framework intended to invite criticism, prior-art corrections, reference implementations, and experimental validation.
Many modern robot-learning systems are implicitly framed as:
perception -> learned policy / RL / VLA / world model -> actuators
HRCA argues that this is the wrong foundation for embodied robots operating near people, animals, fragile objects, equipment, and uncertain environments. AI systems are valuable, but they should not own the body directly.
HRCA proposes a two-part, three-layer, four-block framework.
Layer 1: Harm-Reflex Constraint Layer highest priority
Layer 2: Neuromodulated Motor-Primitive Layer second priority
Layer 3: RL / AI Strategy Layer third priority
The same three-layer constitution is mapped differently onto manipulators, mobile robots, drones, quadrupeds, underwater robots, snake-like robots, humanoids, and other embodiments. Humanoids are the most complex special case, not the canonical case.
The fourth conceptual block, Morphology-Specific Artificial Nervous Body, is not a fourth priority layer. It is the embodiment-specific organization of Layers 1–3.
Harm Reflex governs.
Motor primitives execute.
AI strategy proposes.
Layer 3 can select primitives and propose modulation; it cannot directly bypass Layer 2 or Layer 1. Layer 2 executes body-level closed-loop behavior; it cannot suppress Layer 1. Layer 1 has constitutional priority over both.
flowchart TB
WORLD["Physical world: humans, animals, robot body, task objects, equipment, environment"]
subgraph SENSE["Multi-modal sensing"]
S1["Proprioception"]
S2["Tactile and e-skin"]
S3["Vision, LiDAR, audio"]
S4["Thermal, current, battery"]
S5["IMU and vestibular-like state"]
end
subgraph L1["Layer 1: Harm-Reflex Constraint"]
H1["Harm ontology and risk estimator"]
H2["Reflex library"]
H3["Veto, suppress, rewrite"]
end
subgraph L2["Layer 2: Neuromodulated Motor Primitives"]
P1["Primitive bank"]
P2["Modulator-state vector"]
P3["Local feedback loops"]
end
subgraph L3["Layer 3: RL and AI Strategy"]
A1["Goals and task plans"]
A2["Primitive selection"]
A3["Parameter proposals"]
A4["Long-horizon learning"]
end
ARB["Arbitration Plane"]
LEDGER["Reflex Event Ledger"]
ACT["Actuators"]
WORLD --> SENSE
SENSE --> L1
SENSE --> L2
SENSE --> L3
L3 -->|"selects and modulates"| L2
L3 -.->|"requests; cannot bypass"| L1
L2 -->|"nominal command"| ARB
L1 -->|"constraints and reflex authority"| ARB
ARB --> ACT
ARB --> LEDGER
ACT --> WORLD
HRCA is a design framework for organizing robot control authority:
| Block | Role | Priority meaning |
|---|---|---|
| Harm-Reflex Constraint Layer | Prevent, suppress, interrupt, or rewrite harmful states | Constitutional priority |
| Neuromodulated Motor-Primitive Layer | Execute body-level reusable primitives | Second priority |
| RL / AI Strategy Layer | Plan, learn, select, compose, and modulate | Third priority |
| Morphology-Specific Artificial Nervous Body | Map the three layers into a particular embodiment | Not a priority layer |
HRCA uses cross-cutting mechanisms to make the architecture operational: Arbitration Plane, Harm Ontology, Modulator-State Vector, Primitive Contract Interface, Reflex Event Ledger, Deployment Policy, and Assurance Case.
HRCA does not claim to invent robot pain, robot reflexes, electronic skin, nociceptive sensors, central pattern generators, dynamic movement primitives, impedance control, safe reinforcement learning, shielded reinforcement learning, control barrier functions, runtime assurance, behavior trees, subsumption architecture, whole-body control, neuromorphic robotics, artificial synapses, or robot safety standards.
The term artificial Robot Nervous System (aRNS) is already used in the literature by Kuehn and Haddadin for robot pain and reflexive reaction to potentially damaging contacts. HRCA therefore avoids using aRNS as its name and positions itself as a broader architecture-level systematization.
Subject to further prior-art and legal review, HRCA may cautiously claim a framework-level contribution:
- A unified multi-stakeholder harm formulation: human, animal, robot self, environment, task object, critical equipment, future task capability.
- Harm-reflex as constitutional priority, not auxiliary safety.
- The explicit priority chain: Harm Reflex > Neuromodulated Motor Primitives > RL / AI Strategy.
- AI strategy as planner / selector / modulator, not low-level actuator foundation.
- Morphology-specific artificial nervous body as embodiment-specific organization of the three layers.
- Humanoids as the most complex special case, not the canonical case.
- It does not certify safety.
- It does not replace ISO, ANSI, IEC, or regional robot safety standards.
- It does not certify any implementation.
- It does not prove mathematical optimality.
- It does not solve humanoid robotics.
- It does not claim component-level invention where strong prior art exists.
Recommended reading order:
WHITE_PAPER.mddocs/01-architecture.mddocs/02-priority-arbitration.mddocs/12-cross-cutting-mechanisms.mddocs/06-prior-art-and-claim-boundaries.mddocs/13-research-agenda-and-alternative-architectures.md
hrca/
├── README.md
├── WHITE_PAPER.md
├── RELEASE_INDEX.md
├── RELEASE_MANIFEST.md
├── RELEASE_NOTES_v0.2.1.md
├── CITATION.cff
├── LICENSE
├── LICENSE_RECOMMENDATION.md
├── REFERENCES.bib
├── docs/
│ ├── 00-naming-and-positioning.md
│ ├── 01-architecture.md
│ ├── 02-priority-arbitration.md
│ ├── 03-harm-classes.md
│ ├── 04-modulator-states.md
│ ├── 05-embodiment-mapping.md
│ ├── 06-prior-art-and-claim-boundaries.md
│ ├── 07-experimental-protocols.md
│ ├── 09-x-publication-pack.md
│ ├── 12-cross-cutting-mechanisms.md
│ ├── 13-research-agenda-and-alternative-architectures.md
│ └── 14-bilingual-glossary-and-cn-index.md
├── README.zh-CN.md
├── WHITE_PAPER.zh-CN.md
├── diagrams/
├── spec/
├── experiments/
├── eval/
└── reference/
| Phase | Goal |
|---|---|
| v0.2.1 | Public framework release: architecture, bilingual docs, specs, experiments, prior-art boundaries |
| v0.3 | PDF white paper, expanded reference implementation |
| v0.4 | Minimal simulation benchmark for e-skin / manipulator reflex experiments |
| v0.5 | ROS 2 skeleton for Arbitration Plane and Reflex Event Ledger |
| v1.0 | Reproducible experimental suite across at least two embodiments |
Recommended configuration:
- Documentation, diagrams, and prose: CC BY 4.0
- Schemas and reference skeletons: Apache-2.0
- Bibliographic data: CC0 1.0
See LICENSE and LICENSE_RECOMMENDATION.md.
@techreport{hrca2026,
author = {Hall, Justin},
title = {{HRCA}: Harm-Reflex Constitutional Architecture --
An Artificial Nervous Body Framework for Embodied Robots},
type = {White paper},
number = {v0.2.1},
year = {2026},
url = {https://github.com/genforAI/hrca},
note = {Public design framework. Not a completed product.}
}Contributions are welcome in five areas:
- Prior-art corrections.
- Architecture critique.
- Harm-class and modulator-state refinement.
- Experimental protocols and benchmark scenarios.
- Reference implementation skeletons.
Please do not submit claims that HRCA invented established components such as robot pain, e-skin, CPGs, DMPs, impedance control, safe RL, or runtime assurance. The framework is explicitly prior-art-aware.
HRCA is an architecture proposal. It does not certify safety, authorize deployment, or replace risk assessment, standards compliance, independent verification, or professional engineering judgment.
README.zh-CN.md— Chinese README.WHITE_PAPER.zh-CN.md— Chinese white paper companion.docs/14-bilingual-glossary-and-cn-index.md— bilingual terminology and CN file index.