← Back to Portfolio

VLA's for Autonomous Mobile Navigation

📅 11/2025 - Present
🏫 Stanford University
In Progress

Overview

This research project focuses on advancing autonomous mobile navigation for quadrupedal robots in demanding environments, specifically for search-and-rescue missions. The core of the work involves developing and leveraging Vision-Language-Action (VLA) models to create robust, high-level policy controllers.


By integrating the perceptual richness of VLA models, the robot will move beyond traditional path planning to achieve more sophisticated, human-interpretable navigation and decision-making in complex, unstructured, and often hazardous settings. The exact direction of this project is still being shaped, with ongoing exploration into various VLA extensions.

Key Focus Areas

Technologies & Tools

PyTorch ROS 2 Vision Transformers Reinforcement Learning Isaac Sim/Lab Model Predictive Control LLMs Python C++

Team & Supervision

Fabio Hübel

Researcher
Robotic Systems Lab, ETH Zürich

Dr. Jonas Frey

Research Supervisor
Autonomous Systems Lab, Stanford University

Pascal Roth

Research Supervisor
Robotic Systems Lab, ETH Zürich

Professor Marco Pavone

Research Supervisor
Autonomous Systems Lab, Stanford University

Professor Marco Hutter

Research Supervisor
Robotic Systems Lab, ETH Zürich

Project Gallery

Project visualization 1
[Coming Soon]
Project visualization 2
[Coming Soon]

Results & Impact

Current Status: The project has started very recently, with initial efforts focused on literature review, defining the research scope, and setting up the necessary simulation environments.


Expected Impact: This work aims to significantly advance the field of autonomous mobile robotics by enabling more capable and reliable systems for critical applications like search-and-rescue operations. The integration of language understanding with robotic control represents a major step toward more human-robot collaboration.

References