Conference Paper

Online Speed Adaptation Using Supervised Learning for High-Speed, Off-Road Autonomous Driving.

Conference: IJCAI 2007, Proceedings of the 20th International Joint Conference on Artificial Intelligence, Hyderabad, India, January 6-12, 2007
Source: DBLP


The mobile robotics community has traditionally addressed motion planning and navigation in terms of steering decisions. However, selecting the best speed is also important - beyond its relationship to stopping distance and lateral maneuverability. Consider a high-speed (35 mph) autonomous vehi- cle driving off-road through challenging desert ter- rain. The vehicle should drive slowly on terrain that poses substantial risk. However, it should not daw- dle on safe terrain. In this paper we address one aspect of risk - shock to the vehicle. We present an algorithm for trading-off shock and speed in real- time and without human intervention. The trade-off is optimized using supervised learning to match hu- man driving. The learning process is essential due to the discontinuous and spatially correlated nature of the control problem - classical techniques do not directly apply. We evaluate performance over hun- dreds of miles of autonomous driving, including performance during the 2005 DARPA Grand Chal- lenge. This approach was the deciding factor in our vehicle's speed for nearly 20% of the DARPA com- petition - more than any other constraint except the DARPA-imposed speed limits - and resulted in the fastest finishing time. In mobile robotics, motion planning and navigation have tra- ditionally focused on steering decisions. This paper presents speed decisions as another crucial part of planning - beyond the relationship of speed to obstacle avoidance concerns, such as stopping distance and lateral maneuverability. Consider a high-speed (35 mph) autonomous vehicle driving off-road through challenging desert terrain. We want the vehicle to drive slower on more dangerous terrain. However, we also want to minimize completion time. Thus, the robot must trade-off speed and risk in real-time. This is a natural pro- cess for human drivers, but it is not at all trivial to endow a robot with this ability. We address this trade-off for one component of risk: the shock the vehicle experiences. Minimizing shock is impor- tant for several reasons. First, shock increases the risk of damage to the vehicle, its mechanical actuators, and its elec- tronic components. Second, a key perceptive technology, laser range scanning, relies on accurate estimation of orien- tation. Shock causes the vehicle to shake violently, making accurate estimates difficult. Third, shocks substantially re- duce traction during oscillations. Finally, we demonstrate that shock is strongly correlated with speed and, independently, with subjectively difficult terrain. That is, minimizing shock implies slowing on challenging roads when necessary - a cru- cial behavior to mitigate risk to the vehicle. Our algorithm uses the linear relationship between shock and speed which we derive analytically. The algorithm has three states. First, the vehicle drives at the maximum allowed speed until a shock threshold is exceeded. Second, the vehicle slows immediately to bring itself within the shock threshold using the relationship between speed and shock. Finally, the

Download full-text


Available from: Gabriel Hoffmann,
6 Reads
  • Source
    • "We proposed a method that uses a support vector machine [9] [10]. Stavens et al. focus on assessing the roughness of the terrain instead of grouping the ground surface into classes [11]. There also exist systems that combine vision and vibration sensing [12] [13] [14]. "
    [Show abstract] [Hide abstract]
    ABSTRACT: In outdoor environments, a great variety of ground surfaces exists. To ensure safe navigation, a mobile robot should be able to identify the current ter-rain so that it can adapt its driving style. If the robot navigates in known environ-ments, a terrain classification method can be trained on the expected terrain classes in advance. However, if the robot is to explore previously unseen areas, it may face terrain types that it has not been trained to recognize. In this paper, we present a vibration-based terrain classification system that uses novelty detection based on Gaussian mixture models to detect if the robot traverses an unknown terrain class. If the robot has collected a sufficient number of examples of the unknown class, the new terrain class is added to the classification model online. Our experiments show that the classification performance of the automatically learned model is only slightly worse than the performance of a classifier that knows all classes before-hand.
  • Source
    • "In [20], we suggested an approach that uses Support Vector Machines (SVM) for classification. Stavens et al. presented an approach for vehicles driving up to 35 mph [17]. However, they focused on assessing the roughness of the terrain to adapt the velocity, and not on grouping the ground surface into classes. "
    Proceedings of the 3rd European Conference on Mobile Robots, EMCR 2007, September 19-21, 2007, Freiburg, Germany; 01/2007
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Outdoor robots are faced with a variety of terrain types each possessing different characteristics. To ensure a safe traversal a robot has to infer the current ground surface from sensor readings. Recent techniques generate a model which predicts the terrain class from single vibration signals disregarding the temporal coherence between consecutive mea- surements. In this paper, we present a novel approach in which the final classification relies on the analysis of not only one, but several recent observations. Therefore, the probabilistic framework of the Bayes filter is adopted to the problem of terrain classification. We propose an adaptive approach which automatically adjusts its parameters according to the history of observations. To demonstrate the performance of our method we further describe and compare another technique based on temporal coherence. The evaluation using data collected from our RWI ATRV-Jr robot shows that our approach is both reactive and stable enough to detect fast terrain transitions and selective misclassifications. originally proposed in (6). They showed that vibration sig- natures provide enough information to distinguish between different terrain classes. Usually, accelerometers are used to record vibration data during the robot traversal. The sensors can be attached at the wheels, the axes or the body of the robot. Several researchers considered terrain classification based on vibration data as an instance of a signal processing task. The methods are usually divided into an offline training and an online test or recall phase. In the training phase a model learns to recognize distinct terrain types from labeled vibration patterns. In the recall phase vibration signals of unknown terrain type can be quickly classified using the generated model. In the terrain classification approach of (7) and (8), the model was represented by probabilistic neural networks. While the former approach focused on terrain classifica- tion for electric powered wheelchairs at speeds of 1 and 2m/s, the latter approach considered both a slowly moving autonomous ground vehicle driving at 0.5 and 1m/s and an experimental unmanned vehicle at speeds between 5 and 20mph. Another classification technique was presented by (9). They used principle component analysis for both feature extraction and dimensionality reduction. In a second step, the PCA transformation coefficients were adopted to establish a manifold curve representing the terrain behavior at varying speeds. In a previous paper, we experimentally investigated the performance of several feature extraction schemes (10). These techniques were applied to vibration data recorded at driving speeds between 0.2 and 1.0m/s. The extracted features constituted the inputs for the terrain classification model represented by a support vector machine (SVM). Further research involved the comparison between different classification techniques like the Na¨ive Bayes or the k-Nearest Neighbor classifier (11). It turned out that the SVM outperformed all other techniques with respect to classification performance. In this paper, we propose an extension of the work presented in (10). To motivate our research, we first note that none of the previously mentioned approaches makes use of temporal coherence. Each terrain classification step only considers the current observation. It is very likely, however, that the robot traverses the same terrain type for several measurements. Hence, not only current sensor readings should influence the classification, but also past ones. The model which considers temporal coherence must have the characteristics to be responsive to terrain changes at high frequency. As we will see later, basic approaches fulfill these requirements
    Autonome Mobile Systeme 2007, 20. Fachgespräch, Kaiserslautern, 18./19. Oktober 2007; 01/2007
Show more