Vision-based motion planning for an autonomous motorcycle on ill-structured roads

Abstract

We report our development of a vision-based motion planning system for an autonomous motorcycle designed for desert terrain, where uniform road surface and lane markings are not present. The motion planning is based on a vision vector space (V 2 -Space), which is a unitary vector set that represents local collision-free directions in the image coordinate system. The V 2 -Space is constructed by extracting the vectors based on the similarity of adjacent pixels, which captures both the color information and the directional information from prior vehicle tire tracks and pedestrian footsteps. We report how the V 2 -Space is constructed to reduce the impact of varying lighting conditions in outdoor environments. We also show how the V 2 -Space can be used to incorporate vehicle kinematic, dynamic, and time-delay constraints in motion planning to fit the highly dynamic requirements of the motorcycle. The combined algorithm of the V 2 -Space construction and the motion planning runs in O(n) time, where n is the number of pixels in the captured image. Experiments show that our algorithm outputs correct robot motion commands more than 90% of the time.

Toward featureless visual navigation: Simultaneous localization and planar surface extraction using motion vectors in video streams

“…Using a vector-field approach [20], we develop a lightweight visual navigation algorithm for an autonomous motorcycle. We also have attempted different features for visual odometry such as vertical line segments [21] and high level features [22], [23] to improve robustness.…”

On the error analysis of vertical line pair-based monocular visual odometry in urban area

“…Edges can be easily extracted from images using edge detectors [4]. In visionbased navigation, edge features are widely used in lane detection [5], road following [6], road edge detection [7], and vehicle motion planning [8], [9]. Inspired by the success of applications in different domains, we zoom in vertical lines, which are a very unique set of edges that exist in urban applications.…”

“…3 illustrates the geometric approach. Let l 1 be the line described by (9), which intersects with x-axis at x = b 1 with angle α 1 . Recalling a 1 defined in (9), we have tan k+1) .…”

On the Analysis of the Depth Error on the Road Plane for Monocular Vision-Based Robot Navigation

“…Due to the inherent difficulty in understanding the environment using monocular vision, many researchers focus on applying machine learning techniques to assist navigation [9][10][11][12]. However, those methods are appearance-based and only utilize color and texture information.…”

“…The untrusted area is the region that satisfies (20) and x w < µ d1 . To compute the intersection, we need to understand the relationship between the solution in (20) and the coefficients in (12). Combining them, we know,…”