ROMBUS

Problem

Autonomous Robotics are a rapidly emerging solution in many spaces, but the current computing architecture required to deploy them is not feasible for many users. Large Language Models are commonly used but are too resource intensive, and Small Language Models currently lack a framework to be utilized in robotics.

Approach

The team integrated Small Language Models (SLMs) with robotic systems for natural language task planning. They enhanced precision using Vision Language Models and developed a benchmarking framework to measure efficiency and accuracy for different SLMs.

Solution

ROMBUS enables low-latency execution of robotic tasks using SLMs, converting natural language commands into actions. It includes a benchmarking framework to evaluate SLM performance and supports real-time automation with optimized perception and control. This allows performing limited types of robotics tasks without the large resources required for LLM deployment, allowing for more efficient use of robotics solutions.

Team Members

Lakshita Singh

Aayush Kumar

Kendra Yang

Matthew Zhang

View the team's poster here (PDF)

Back to Projects

Share