Legged locomotion endows quadruped robots with exceptional flexibility and adaptability in unstructured, cluttered environments [1]. One of their most compelling applications is autonomous navigation through such complex spaces. Successful navigation requires an integrated pipeline of perception, path planning, and control. In this project, you will develop and validate a navigation pipeline:

1. Perception: Fuse sensor data (e.g., vision and lidar) to build a real-time, obstacle-aware map of an indoor navigation environment [2].

2. Path Planning: Implement a planner that computes safe, dynamically feasible trajectories through clutter—accounting for both geometric obstacles and the robot’s kinematic constraints.

3. Control: Implement a low-level controller to track planned trajectories robustly, handling model uncertainties and small disturbances.

You will first integrate and test the full stack in simulation (e.g., using Nvidia Issac Sim [3]). After that, you will deploy it on a Unitree GO2 [4] quadruped and demonstrate autonomous traversal of a real office environment, evaluating performance metrics such as speed, success rate, and energy efficiency.

 

Main Objectives:

 

• Literature research on quadruped navigation

• Setup a navigation scenario in Nvidia Issac Sim

• Implement the navigation pipeline and test in simulation

• Deploy the implemented pipeline on a real quadruped robot and Test in a real world navigation scenario.

 

Deliverables:

• Clean and optimized workflow for quadruped robot navigation in simulation and real world

• Reusable and documented code base

• Master thesis

 

[1] Learning a State Representation and Navigation in Cluttered and Dynamic Environments arxiv.org/pdf/2103.04351

[2] github.com/nvidia-isaac/nvblox

[3] developer.nvidia.com/isaac/sim

[4] www.docs.quadruped.de/projects/go2/html/go2_driver.html

Mandatory:

• Strong Programming skills in Python

• Proficient background in robot motion planning and control

• Proficient background in SLAM

• Background in Linux system

• Background in robot simulation, such as Issac

 

Optional:

• Familiar with ROS2

• Familiar with Deep learning frameworks, such as Tensorflow, Pytorch