[Paper Review] MuSHR: A Low-Cost, Open-Source Robotic Racecar for Education and Research

Siddhartha S Srinivasa, Patrick Lancaster+9 more|arXiv (Cornell University)|Aug 21, 2019

Robotic Path Planning Algorithms2 references45 citations

TL;DR

MuSHR is a low-cost, open-source robotic racecar platform for education and research, featuring a full-stack hardware/software stack and detailed tutorials to enable broad access.

ABSTRACT

We present MuSHR, the Multi-agent System for non-Holonomic Racing. MuSHR is a low-cost, open-source robotic racecar platform for education and research, developed by the Personal Robotics Lab in the Paul G. Allen School of Computer Science & Engineering at the University of Washington. MuSHR aspires to contribute towards democratizing the field of robotics as a low-cost platform that can be built and deployed by following detailed, open documentation and do-it-yourself tutorials. A set of demos and lab assignments developed for the Mobile Robots course at the University of Washington provide guided, hands-on experience with the platform, and milestones for further development. MuSHR is a valuable asset for academic research labs, robotics instructors, and robotics enthusiasts.

Motivation & Objective

• Democratize robotics by providing an affordable, open, and easily buildable racecar platform for education and research.
• Offer a complete end-to-end hardware/software stack with detailed documentation and tutorials.
• Support university courses and research through ready-to-run localization, planning, and control modules.
• Enable rapid prototyping and experimentation in multi-robot and autonomous navigation contexts.

Proposed method

• Hardware built from off-the-shelf components with detailed assembly instructions.
• A four-sensor perception stack including RGBD camera (Intel RealSense D435i) and LIDAR (YLDAR X4).
• A Nvidia Jetson Nano for onboard computation with power/subsystems suitable for typical RC batteries.
• Software stack with four main components: sensing interface, control module, ESC interface, and localization module.
• An autonomous model predictive controller (mushr_rhc) capable of static/dynamic trajectory planning with map-based obstacle avoidance.
• A localization module based on a particle filter adapted from prior work, integrated into a ROS-based navigation stack.
• Open-source ROS interfaces for sensors, and teleoperation support via a game controller, with safety/backup controls.

Experimental results

Research questions

• RQ1 How affordable is a fully functional autonomous racecar platform suitable for education and research?
• RQ2 Can a low-cost stack support localization, planning, and control algorithms comparable to higher-cost counterparts?
• RQ3 How effective is the MuSHR platform as a teaching tool in university courses for mobile robots, localization, and autonomous navigation?
• RQ4 What is the impact of open documentation and tutorials on enabling broad adoption and contribution to the platform?

Key findings

• MuSHR can be built for $600 baseline, and about $900 with laser scanner and RGBD camera, significantly cheaper than comparable MIT-based setups.
• The platform uses a Nvidia Jetson Nano, which enables running localization, planning, and machine learning tasks at lower cost.
• MuSHR includes a four-sensor sensing stack (RGBD camera, LIDAR, bumper) with ROS interfaces and teleoperation support.
• A four-component software stack enables sensing, control (including an autonomous model predictive controller), ESC interfacing, and localization via a particle filter.
• Open, extensive documentation and video-based build guides are provided, supporting rapid deployment and education.
• MuSHR has been integrated into UW courses (CSE 490R, CSE 571, EE P 545) and used for experiments in localization, control, and planning.
• The platform aims to enable education and research from high-school to university level with ongoing tutorials and community-driven development.