Optimal Sensor Setup for USV Autonomy in Obstacle Navigation

Build
1

Hello everyone,

My team and I are developing an unmanned surface vehicle (USV) for a university competition. Our USV is a catamaran equipped with a Jetson Nano for processing. The total budget for our vehicle, including all components, is $1900.

Competition Task

The goal of the competition is to autonomously navigate through a parkour with various obstacles. The USV must:

  • Detect and classify obstacles,
  • Plan a route around them,
  • Complete the parkour as quickly as possible.

I have attached a file with the parkour details for better context.

parkour
parkour1730×647 108 KB

Sensor Setup Options

We are evaluating different sensor configurations and would like to hear your opinions on their feasibility and efficiency:

  1. Monocular vision + LIDAR
  2. Monocular vision + AI-based depth estimation + LIDAR
  3. Stereo vision
  4. Stereo vision + LIDAR

Our primary concerns are accuracy and cost-effectiveness. We plan to develop accuracy enhancement, prediction, and verification algorithms to improve sensor performance. However, I have no prior experience with LIDAR, so I am unsure about its real-world accuracy in a marine environment.

Given our budget, processing limitations (Jetson Nano), and the need for fast and reliable obstacle detection, which sensor setup would you recommend? Any insights from your experience would be greatly appreciated!

Looking forward to your responses.

Best regards.

2

Hi @UmutCee -
ArduRover, used on the BlueBoat, already supports obstacle avoidance to a limited extent - it may be a good place to start from! You’d just need one of the supported sensors listed…

StereoVision is cheaper than Lidar typically - an extension for BlueOS that supports the OakD lite stereo camera already exists! We’ve experimented with some Lidar systems as well, but generally in larger sea states they can detect the ocean itself, which can cause problems…

It sounds like a fun project, best of luck!

3

Thank you for your response!

I wasn’t aware that ArduPilot has an obstacle avoidance mode—this is great to know. The area where we’ll be operating is mostly a calm, waveless sea, and I don’t expect the wave height to exceed our vehicle’s height. To prevent a 2D LiDAR from detecting the sea surface as an obstacle, I believe I’ll need to mount it on a gimbal system to keep it parallel to the water at all times.