Uncertainty-Aware Occupancy Map Prediction Using Generative Networks for Robot Navigation

Abstract

Fast, collision-free motion through unknown environments remains a challenging problem for robotic systems. In these situations, the robot's ability to reason about its future motion is often severely limited by sensor field of view (FOV). By contrast, biological systems routinely make decisions by taking into consideration what might exist beyond their FOV based on prior experience. In this paper, we present an approach for predicting occupancy map representations of sensor data for future robot motions using deep neural networks. We evaluate several deep network architectures, including purely generative and adversarial models. Testing on both simulated and real environments we demonstrated performance both qualitatively and quantitatively, with SSIM similarity measure up to 0.899. We showed that it is possible to make predictions about occupied space beyond the physical robot's FOV from simulated training data. In the future, this method will allow robots to navigate through unknown environments in a faster, safer manner.

Efficient Map Prediction via Low-Rank Matrix Completion

“…Map extrapolation attracts increasing attention in recent years and almost all the current methods for map extrapolation are based on deep neural networks (DNNs) [8], [9], [10], [11], [12], [13]. However, DNNs based methods require huge amount of data for training and hence are still limited in general applications.…”

Exploration and Mapping with Groups of Robots: Recent Trends

“…For example, structures that repeat in the environment are used to predict potential geometries and loop closures in the unknown part of the environment [61,62]. More recent work leverage the success of convolutional neural networks [63•] and generative networks [64,65] trained on datasets composed of occupancy grid maps to perform predictions on unknown parts of the environment. Map inference can then be used to estimate more accurately the value of information gain and used to calculate the utility of the next frontiers.…”

Parameter Reduction and Optimisation for Point Cloud and Occupancy Mapping Algorithms

“…In robotics, occupancy maps have a wide range of applications, including spatial representation of the real world [1], navigation [2], motion planning [3] and autonomous driving [4]. Maps are commonly generated from point clouds with a variety of sensors such as LIDAR [5], RGB-D cameras [6] and stereo cameras [7].…”

“…• 1 is the first parameter for disparity smoothness control. A reasonably good sample in OpenCV is 1 = 8 2 , where is the number of image channels. • 2 is the second parameter for disparity smoothness control.…”