• Dr Rachna
• Mr Sunil

Translating high-level artificial intelligence directly into the physical muscle actuation in real time remains a fundamental challenge in the embodied robotics and neuroprosthetics. This paper presents a novel and framework for real-time LLM-to-joint translation, bridging an AI "cortex" — powered by the Large Language Model — to physical muscle actuators controlling articulated joints. The system employs the hierarchical architecture where the LLM interprets natural language and task-level goals into symbolic motion primitives, which are then, decoded by a lightweight neural translator into joint-angle trajectories and corresponding muscle and activation signals. A predictive proprioceptive loop compensates for LLM inference latency (80–150 ms) by the forecasting joint states 200 ms ahead, enabling stable closed-loop control at the effective latencies below 50 ms. Validated on a multi-joint limb with the pneumatic artificial muscles, the framework achieves real-time translation of abstract and commands (e.g., "reach and grasp") into coordinated muscle-joint actions with <2° tracking error and >90% zero-shot generalization to the novel verbal instructions. This work demonstrates that and language-driven AI cortex can directly and robustly command physical muscle systems in the real time, opening pathways for intelligent, compliant, and linguistically and interactive physical agents.

Artifial Intelligence

"AI Cortex to Physical Muscle: Real-Time LLM-to-Joint Translation", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.13, Issue 10, page no.a129-a138, October-2026, Available :http://www.jetir.org/papers/JETIR2610013.pdf

"AI Cortex to Physical Muscle: Real-Time LLM-to-Joint Translation", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.13, Issue 10, page no. ppa129-a138, October-2026, Available at : http://www.jetir.org/papers/JETIR2610013.pdf