Work

Introducing JARVIS: An LLM based Autonomous Robotic Agent for Near Real-Time Planning and Execution of Novel ADL Tasks

Year: 2026

Type: conference-paper

Abstract: Motivated by the need for robots to perform complex multistep tasks many of which are novel and unique to the needs of people with disabilities using an assistive robot, this paper designs an LLM-based pipeline that starts with natural language prompts and terminates in usable code for a robot.The only code designed and implemented by a programmer is a small fixed dictionary of motion and/or force control primitives that the LLM exploits repeatedly to create novel tasks from scratch.This paper details the current status and future plans for JARVIS, the LLM-based Agent, and the results showcase the flexible interaction of the agent with the user to reliably complete several novel Activities of Daily Living (ADL) tasks.

Source: Proceedings of the International Conference of Control, Dynamic systems, and Robotics

DOI: 10.11159/cdsr26.140

Authors: Ronald Fuller, Nihad Habizada, Aman Behal

Institution: University of Central Florida

Language: English

Topic: Robot Manipulation and Learning

Subfield: Control and Systems Engineering

Field: Engineering

Domain: Physical Sciences

Open Access status: bronze

Funders: Australian Government, National Institute on Disability, Independent Living, and Rehabilitation Research

Awards: G3513040557, G8047338286