TRAVERSABILITY PREDICTION FOR UNMANNED GROUND VEHICLES BASED ON IDENTIFIED SOIL PARAMETERS Suksun Hutangkabodee, Yahya H Zweiri, Lakmal D Seneviratne, Kaspar Althoefer Department of Mechanical Engineering, King’s College London Strand, London WC2R 2LS, UK {suksun.hutangkabodee, yahya.zweiri, lakmal.seneviratne, k.althoefer}@kcl.ac.uk Abstract: A novel technique for identifying soil parameters on-line while traversing with a tracked vehicle on unknown terrain is presented. This technique, based on the Newton Raphson method is used to identify unknown soil parameters. Comparing with the Least Square method, it shows that the Newton Raphson method is better in terms of prediction accuracy, computational speed, and robustness to initial conditions and noise. For heavy tracked vehicle, cohesion has negligible effect on the vehicle performance. These identified soil parameters are then employed for traversability prediction for a tracked vehicle travelling on unknown terrain. Copyright © 2005 IFAC Keywords: Parameter Identification, Tracked Vehicles, Traversability, Interaction Dynamics. 1. INTRODUCTION Unmanned Ground Vehicles (UGVs) have many potential applications, including space exploration, defense, agriculture, mining and construction. Most unmanned ground vehicles are currently controlled by tele-operation. Tele-operation requires continuous and repetitive human intervention, which hampers the speed of the vehicle and the range of potential applications (Zweiri et al., 2003). Further they have problems due to bandwidth limitations and communication time delays of the transmission link. Increased autonomy of ground vehicles will not only improve the safety of the operators, but also increase the range of potential applications. Research on the autonomy aspect of UGV has been carried out in the past. Model-based autonomy is described by Kurien et al. (1998). This approach involves the use of automated reasoning algorithm and first principles models of physical system being controlled to achieve robust and autonomous operation, even in failures or anomalous situations. It is now being applied and developed for the NASA Mars Mission. Behavior-based autonomy is presented by Langer et al. (1994) and Rosenblatt (1997). In both papers, Distributed Architecture for Mobile Navigation (DAMN) is employed as a behavior-based architecture for autonomous mobile navigation system. It is a planning and control architecture in which a collection of independently operating behaviors collectively determines a robot’s actions. Simultaneous Localisation And Mapping (SLAM) is presented by Nieto et al. (2003) and an algorithm named FastSLAM which addresses data association and real time implementation of the SLAM problem from a Bayesian point of view is employed. Dead reckoning estimation is applied by Schnberg et al. (1995). The dead reckoning position estimation and the absolute position measurement are fused by using Kalman filtering techniques to provide a corrected estimate. This approach is used to improve the heading error of an autonomous vehicle. The combination of fuzzy logic and neural network is employed by Freisleben and Kunkelmann (1993). The approach is used to tackle the problem of controlling a car to drive autonomously around an unknown race track. The basic idea of this proposal is to let a fuzzy controller supply the training data for a backpropagation neural network and used the trained network to drive the car on an unknown track race. The autonomy of UGV can not only be improved by various methods mentioned above, but can also be improved by UGV acquiring information from the terrain which the UGV traverses on. This is where