Solving Inverse Kinematics Using Large Language Models - MSc Thesis Proposal by : Steven Rice

Thursday, February 6, 2025 - 09:30

The School of Computer Science is pleased to present…

 

Solving Inverse Kinematics using Large Language Models

MSc Thesis Proposal by: Steven Rice

Date: Thursday, February 6th, 2025

Time:  9:30 AM

Location: Essex Hall, Room 122

 

Abstract:

Inverse Kinematics (IK) is an integral part of robot manipulation. IK can be challenging to be solved by a human, and many computer-aided approaches have been proposed but each has its limitations. The emergence of Large Language Models (LLMs) has seen them applied to solving complex tasks including math problems. This work proposes "LLM-IK"—the first methodology to utilize LLMs to solve IK problems. The methodology consists of extracting relevant serial manipulator information from its configuration details, prompt engineering, and providing the LLMs with methods and feedback to interact with and learn about the kinematic chain it is solving. This allows the LLMs to query information and test its hypotheses, allowing it to get concrete validation alongside their own beliefs and conclusions.

Additionally, multiple methods of extending solutions of sub-chains to solve more complex kinematic chains are tested. This allows for incremental problem solving, allowing the LLMs to approach the complex problem of solving IK into distinct sub-problems. With more powerful LLMs continually being developed, this methodology could be used to produce highly accurate and efficient IK solutions with state-of-the-art performance in the future. Additionally, as analytically solving IK is a complex math problem, this methodology of strategically solving sub-problems along with multiple methods of extending solutions could be applied to other math problems solved by LLMs.

 

Keywords: large language models, inverse kinematics, robotics, prompt engineering, reasoning

 

Thesis Committee:

Internal Reader: Dr. Alioune Ngom

External Reader: Dr. Ahmed Azab Ismail

Advisor: Dr. Sherif Saad Ahmed

Vector Logo

Registration Link (Only MAC Students Need to Pre-Register)