Multimodal terrain analysis for an all-terrain crisis management robot

As the evolution in robotic technology is continuing, robots are more and more leaving the protected lab environment and entering the unstructured and complex outside world, e.g. for applications such as humanitarian demining or more generic crisis management tasks. Contrary to the indoor case, any robotic system which needs to navigate through an unstructured outdoor environment requires a means to judge whether the terrain in front of the robot is traversable or not. This so-called terrain traversability problem is no easy problem, as it depends on multiple factors. Some of these factors are external to the robotic system and very hard to measure or model, e.g. soil, vegetation, rocks, slopes, humidity, ... Other factors are intrinsic to the robotic system used, e.g. the wheel diameter, motor torque, drivetrain system, ... It is clear that a generic terrain traversability estimation methodology, which would aim to model all of the aforementioned influences, would lead to an algorithm with an insurmountable number of difficult-to-tune parameters. Therefore, existing approaches towards terrain traversability estimation aim to simplify the problem. Two general approaches can be distinguished for doing this, each having their advantages and disadvantages. One type of techniques stems from the computer vision community and aims to classify the terrain based upon color image data from an on-board camera. A second series of approaches uses depth information, obtained either using a 3D Laser or a stereo camera, to infer traversability information. In this paper, a novel stereo-based terrain-traversability estimation methodology is proposed. The novelty is that – in contrary to classic depth-based terrain classification algorithms – all the information of the stereo camera system is used, also the color information. Using this approach, depth and color information are fused in order to obtain a higher classification accuracy than is possible with uni-modal techniques.