Learning traversability models for autonomous mobile vehicles

Abstract

Autonomous mobile robots need to adapt their behavior to the terrain over which they drive, and to predict the traversability of the terrain so that they can effectively plan their paths. Such robots usually make use of a set of sensors to investigate the terrain around them and build up an internal representation that enables them to navigate. This paper addresses the question of how to use sensor data to learn properties of the environment and use this knowledge to predict which regions of the environment are traversable. The approach makes use of sensed information from range sensors (stereo or ladar), color cameras, and the vehicle's navigation sensors. Models of terrain regions are learned from subsets of pixels that are selected by projection into a local occupancy grid. The models include color and texture as well as traversability information obtained from an analysis of the range data associated with the pixels. The models are learned without supervision, deriving their properties from the geometry and the appearance of the scene.The models are used to classify color images and assign traversability costs to regions. The classification does not use the range or position information, but only color images. Traversability determined during the model-building phase is stored in the models. This enables classification of regions beyond the range of stereo or ladar using the information in the color images. The paper describes how the models are constructed and maintained, how they are used to classify image regions, and how the system adapts to changing environments. Examples are shown from the implementation of this algorithm in the DARPA Learning Applied to Ground Robots (LAGR) program, and an evaluation of the algorithm against human-provided ground truth is presented.

Terrain Roughness Identification for High-Speed UGVs

“…This allows for a UGV to make navigation decisions using the predicted terrain roughness before physically encountering the terrain, preventing dangerous vehicle-terrain interaction forces. Existing predictive techniques that identify traversable and non-traversable terrains in [8]- [11] do not provide the same level of detail as the RI and allow only for navigation decisions regarding where not to travel. As seen in the results shown in this paper terrain is identified with varying levels of roughness meaning a UGV can make more precise navigation decisions.…”

“…One approach to appearance based terrain identification is to perform binary classification whereby terrain is identified as either traversable or non-traversable [8]- [11]. The approaches by Shneiera et al in [8] and Thrun et al in [9] were in fact a hybrid approach which used geometric sensors to identify nearby non-traversable features with an algorithm identifying similar appearing non-traversable regions in camera images for longer range terrain prediction. Kim et al took a different approach by training a visual classifier to identify nontraversable regions through a hand labelled training set [10].…”

Non-parametric Learning to Aid Path Planning over Slopes

“…This task of interpreting exteroceptive data by associating a scalar value of cost or utility is referred to as Scene Interpretation in this work. In ground vehicle robotics the focus is usually on identifying hard hazards such as obstacles or classifying predefined environmental states into different degrees of traversibility (Ojeda et al 2006;Jackel et al 2006;Hadsell et al 2007;Shneier et al 2008). Assumptions such as terrain homogeneity, or perpetual existence of a road are often made to simplify the problem.…”

A model of computation and representation in the brain

“…It is informed by over 40 years of experience in the design, engineering, and testing of a wide variety of intelligent machine systems that have successfully performed complex tasks in real-world environments. These include sensory-interactive robot manipulators [8], the NBS automated manufacturing research facility (AMRF) [79], control system architectures for automated coal mining systems, automation for post office general mail facilities and stamp distribution centers, intelligent controllers for multiple undersea vehicles, automation systems for next-generation nuclear submarines, enhanced machine tool controllers for automotive and aircraft prototype machining and assembly cells, advanced controllers for commercial water jet cutting machines, and a number of Army Research Lab, DARPA, and Federal Highway Administration research programs for intelligent unmanned ground vehicles [7,2,58,81].…”