Multi-sensor fusion of occupancy grids based on integer arithmetic

Abstract

For the last 25 years, occupancy grids have been intensively used as a well-understood framework for many robotic applications, such as path planning or obstacle avoidance. They offer a unifying framework for multiple heterogeneous sensor integration using a probabilistic representation of sensor data. This integration is computed through Bayesian techniques or evidence combination approaches, both requiring high computation workload using real number representation. In critical application domains, it is challenging to fuse data coming out of several sensors in real-time using constrained embedded platforms. In this paper, we propose a revised theoretical formulation of multi-sensor fusion using only integer arithmetic. We apply this novel framework to compute occupancy grid by only using integer numbers to represent probabilities. Compared to the state-of-the-art solutions, our fusion framework enables implementation on platforms with no floating-point support. Our experiments demonstrate that fusion of real automotive data from a 4-scans LIDAR can be integrated into a microcontroller without a floating-point unit. Our approach opens the perspective for microcontroller or even for hardware block based on ASIC or FPGA to support occupancy grid applications with real-time performance.

DRC2: Dynamically Reconfigurable Computing Circuit based on memory architecture

“…Such grids encode occupation information using signed binary words (e.g. 8-bit words) [21]. The occupation data is then manipulated by the application as a simple mono-channel image buffer.…”

An OGM‐Based Visual Navigation System with Reduced Homographies and Accurate Computations for Mobile Robots

“…In [9], a SLAM task is performed at hardware level by using genetic algorithms and updating just the grid-elements along the exploration area of the sensor. With the purpose of reducing hardware requirements, in [10] there are several sensors integrated that uniquely use integer arithmetic in order to represent the probability of grid-elements. A novel sensor model to map and detect moving objects is proposed in [11].…”

“…Here, environment modeling plays a key role and several methods have been proposed, e.g., Occupancy Grid Mapping (OGM) [3], Geometric Mapping (GM) [4], and Topological Mapping (TM) [5], with the former being the predominant one. The OGM method creates maps from environments which have been sensed by either a sonar [3,6] or a different sensor such as laser range finder (e.g., Laser Imaging Detection and Ranging (LIDAR)) [7][8][9][10][11], ultrasonic [12], radar [13], and vision [14][15][16][17][18], as well as using a combination of them [19][20][21][22][23][24]. Thereby, any environment can be mapped using OGM-based methods, albeit estimating the localization of a robot based on its position and orientation in a mapped environment with a high-resolution may require high computational costs [25].…”

A Comparison of FPGA and GPGPU Designs for Bayesian Occupancy Filters

“…These attempts use floating-point representation for probability estimation and fusion, but highly constrained embedded platforms will not often feature a floating-point unit. A fixed point design of an occupancy grid representation—BOF preprocessing step—was introduced by [ 28 ] to overcome that limitation. It shows good behavior, thereby opening the door to more HW-oriented implementations of the BOF (e.g., using FPGA).…”