Pascal Roth

Pascal Roth

Ph.D. in Robot Learning & Software Engineer

ETH Zurich

NVIDIA

Biography

I am a doctoral researcher at the Robotic Systems Lab, ETH Zurich, supervised by Prof. Marco Hutter, and a Software Engineer at NVIDIA where I am a core developer of Isaac Lab β€” a GPU-accelerated open-source framework for robot learning used across industry and academia.

My research focuses on learning-based navigation and autonomy for legged robots, with particular interest in perceptive planning, forward dynamics models, and sim-to-real transfer. I develop methods that allow robots to reason about physical interactions with complex terrain, enabling safer and more capable autonomous navigation in unstructured environments.

Prior to my PhD, I completed an M.S. in Mechanical Engineering at ETH Zurich (GPA: 5.70/6), where my thesis introduced ViPlanner β€” a visual-semantic local navigation framework demonstrating zero-shot sim-to-real transfer on ANYmal. I also hold a B.S. with Honours from TU Darmstadt.

Interests
  • Robotic Navigation & Autonomy
  • Robot Learning (RL / Imitation)
  • Simulation & Sim-to-Real Transfer
Education
  • PhD in Robot Learning

    ETH Zurich

  • MSc in Mechanical Engineering, 2023

    ETH Zurich

  • BSc in Mechanical Engineering, 2020

    TU Darmstadt

Experience

 
 
 
 
 
NVIDIA Switzerland AG
Software Engineer
January 2025 – Present Zurich
Core developer of Isaac Lab, a GPU-accelerated open-source framework unifying reinforcement learning, imitation learning, and motion planning across diverse robotic platforms.
 
 
 
 
 
Robotic Systems Lab, ETH Zurich
Teaching Assistant β€” Robot Dynamics
January 2024 – Present Zurich
Course design, midterm examination development, and graduate-level instruction for Robot Dynamics at ETH Zurich.
 
 
 
 
 
Robotic Systems Lab, ETH Zurich
Research Engineer
August 2023 – January 2024 Zurich
Research on Forward Dynamics Models for quadrupedal navigation; development of Isaac Lab (formerly ORBIT).
 
 
 
 
 
Ansys Inc.
Research Internship
January 2021 – August 2021 Munich, Germany
Error modeling and compensation using Machine Learning in turbulent Large-Eddy-Simulations.

Education

 
 
 
 
 
Robotic Systems Lab, ETH Zurich
Ph.D. in Robot Learning
February 2024 – Present Zurich
Focus on Robotic Navigation and Autonomy Research. Advisor β€” Prof. Marco Hutter.
 
 
 
 
 
ETH Zurich
M.S. in Mechanical Engineering
September 2021 – July 2023 Zurich
Specialization in robot learning, autonomous navigation, and control. Thesis β€” ViPlanner.
 
 
 
 
 
ETH Zurich
ETH Robotics Summer School
July 2022 – July 2022 Zurich
Search and Rescue competition using an autonomous rough-terrain UGV; built full autonomy stack including state-estimation, SLAM, motion planning, and object detection.
 
 
 
 
 
Tsinghua University
Exchange Semester
September 2019 – January 2020 Beijing, China
 
 
 
 
 
TU Darmstadt
B.S. in Mechanical and Process Engineering
October 2017 – December 2020 Darmstadt, Germany

Recent Publications

(2025). Isaac Lab: A GPU-Accelerated Simulation Framework for Multi-Modal Robot Learning. arXiv.

Code Project

(2025). Risk-Guided Diffusion: Toward Deploying Robot Foundation Models in Space, Where Failure Is Not An Option. arXiv.

PDF

(2024). ETHcavation: A Dataset and Pipeline for Panoptic Scene Understanding and Object Tracking in Dynamic Construction Environments. ACRA.

PDF Cite Code Dataset Project Video

(2024). ViPlanner: Visual Semantic Imperative Learning for Local Navigation. ICRA.

PDF Cite Code Project Video

Projects

*
ViPlanner - Visual Semantic Imperative Learning for Local Navigation
Local Path Planning with semantic and geometric understanding of the environment.
ViPlanner - Visual Semantic Imperative Learning for Local Navigation
Visual Odometry
Monocular continuous Visual Odometry (VO) pipeline designed for the purpose of vehicle localization
Visual Odometry
Orbit
A Unified Simulation Framework for Interactive Robot Learning Environments
Orbit
ETH Robotics Summer School
Search and Rescue Challenge
ETH Robotics Summer School
Self-Supervised Panoptic Segmentation
Self-supervised pre-training strategies for environment-specific segmentation models
Self-Supervised Panoptic Segmentation
Motion Planning
Using RRT* with motion primitives to guide a starship through a map with static and dynamic obstacles.
Motion Planning

Contact