Fusion of Lidar and Stereo Range for Mobile Robots

TLDR

The project aims to fuse lidar and stereo sensors with modular algorithms to produce a unified spatial map of obstacles and free space for a mobile robot. A unified architecture employs a spatial object representation—including an occupancy grid, geometric scaling, density, and confidence estimates—to integrate lidar and stereo data and track obstacles over time. The framework proved flexible, successfully fusing diverse lidar and stereo data with minimal code changes, and automatically compensating for sensor failures, indicating promise as a robust mobile‑robot sensor‑fusion system.

Abstract

This project involves the use of two physical sensors and several modular algorithms to generate a single spatial representation of obstacles and free-space surrounding a mobile robot. This representation is used to track obstacles over time. We deflne a unifled architecture that utilized a spatial object representation as the default communication conduit between all modules. This representation includes an occupancy grid of the area immediately surrounding the robot, along with geometric scaling and density information, and confldence estimates for these data. The ∞exibility of this architecture has been demonstrated by utilizing both live and recorded data sets from difierent lidar sensors, cameras, and processing modules, with minimal or no changes to the processing code. This system shows promise as a ∞exible sensor fusion framework for mobile robotics. Advantages to this system over existing systems include the ability to encapsulate sensors, so that downstream algorithms don’t need to know the details of the sensor suite on the robot, and for the system to automatically adjust for conditions where a single sensor fails or performs poorly but other sensors are functioning properly, such as in open flelds where the range of lidar data is quite limited or in grassy areas where stereo range has wellknown problems.