Agentic Harnesses: LLM-Driven Verification Layers for Robot Autonomy

Source

Source name: arXiv
Source kind: paper
Authors / organization: Rohan Bhagra, Mahantesh Halapannavar, Uddhav Bhattarai
Venue: arXiv: cs.RO
Published: 2026-08-10
Collected: 2026-08-11
URL: http://arxiv.org/abs/2608.09857v1
PDF: https://arxiv.org/pdf/2608.09857v1
DOI: UNKNOWN
Registry group: academic_apis_and_preprints
Evidence role: canonical_academic_index
Language: en
Country / region: UNKNOWN

Plain-Language Brief

Researchers propose a verification layer using large language models (LLMs) to ensure the feasibility and ethical alignment of actions planned by robot autonomy systems. The system can evaluate the permissibility of actions, identify potential risks such as biases or adversarial attacks, and either approve, reject, or escalate tasks for human review.

What the Document Says

The focus of research in robot autonomy has been on execution rather than verifying feasibility of action planning models.
Robotics planning models, like LLMs, can carry risks such as being biased towards user-specified goals, unsafe due to lack of memory about prior safety risks, or vulnerable to adversarial attacks on the autonomy ecosystem.
A proposed solution is a LLM-driven verification layer between planning and execution to evaluate action permissibility.
The LLM-as-a-Judge ensemble uses chain-of-thought reasoning across models and synthesizes outputs from expert judges, mimicking a combination of a mixture of experts and self-consistency approach.
This verification layer acts as middleware, controlling plans from the server's planning module before they reach the MCP server and the robot's low-level controls.
The system achieves near 85% precision across accept/escalate/reject categories with 97% containment of adversarial attacks and negligible errors between accepting and rejecting tasks, with most errors occurring at the escalate boundary.

Key Concepts Mentioned

robot autonomy
verification layer
large language models (LLMs)
action permissibility
chain-of-thought reasoning
adversarial attacks

Claims Made by the Document

Robotics planning models can carry risks.
A LLM-driven verification layer can evaluate action permissibility.
The proposed system achieves near 85% precision and 97% containment of adversarial attacks.

Evidence Used

Advances in advanced artificial intelligence tools have sparked research in robot autonomy, but the development of such systems has largely focused on execution rather than verifying the feasibility actions planning models propose.
Like general-purpose LLMs, robotics planning models carry risks.
We propose a LLM-driven verification layer between planning and execution to evaluate action permissibility.
Our LLM-as-a-Judge ensemble combines chain-of-thought reasoning across models and synthesizes those expert judge outputs.
This layer serves as middleware, gating plans from the server's planning module before they reach the MCP server and therefore the robot's low-level controls.
With this system, we achieve near 85% precision across accept/escalate/reject categories 97% containment of adversarial attacks, with negligible errors between accepting and rejecting tasks, and errors mostly manifesting at the escalate boundary.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Advances in advanced artificial intelligence tools have sparked research in robot autonomy, but the development of such systems has largely focused on execution rather than verifying the feasibility actions planning models propose. Like general-purpose LLMs, robotics planning models carry risks: biased toward user-specified goals, they may suggest actions misaligned with scientific ethics, they may be unsafe due to an inability to 'remember' prior safety risks, or they may be vulnerable to adversarial attacks on the autonomy ecosystem. We propose a LLM-driven verification layer between planning and execution to evaluate action permissibility. Our LLM-as-a-Judge ensemble combines chain-of-thought reasoning across models and synthesizes those expert judge outputs, mirroring a combination of a mixture of experts and self-consistency approach. This layer serves as middleware, gating plans from the server's planning module before they reach the MCP server and therefore the robot's low-level controls: plans are approved, rejected for reformulation, or escalated for human review. With this system, we achieve near 85% precision across accept/escalate/reject categories 97% containment of adversarial attacks, with negligible errors between accepting and rejecting tasks, and errors mostly manifesting at the escalate boundary.