Efficient Training of Artificial Neural Networks for Autonomous Navigation

Abstract

Many real world problems quire a degree of flexibility that is diflicult to achieve using hand programmed algorithms. One such domain is vision-based autonomous driving. In this task, the dual challenges of a constantly changing environment coupled with a real t h e processing constrain make the flexibility and efficiency of a machine learning system essential. This chapter describes just such a learning system, called ALVINN (Autonomous Land Vehicle In a Neural Network). It presents the neural network architecture and training techniques that allow &VI" to drive in a variety of circumstanm including singlelane paved and unpaved roads, multilane lined and unlined roads, and obstacle-ridden on-and offroad environments, at speeds of up to 55 miles per hour.

Vision for mobile robot navigation: a survey

“…The NAVLAB 1 vehicle was also equipped with a neuralnetwork based navigational system, called ALVINN [124], [125], [126], [127]. The ALVINN system (Autonomous Land Vehicle In A Neural Network) was first reported in 1989 [124].…”

“…Starting in 1986, when a Chevy van was converted into NAVLAB 1, until today's converted metro buses known as Navlab 9 and 10, CMU and its partners have developed a family of systems for automated navigation on highways. Four of these systems are: RALPH (Rapidly Adapting Lateral Position Handler) [128], ALVINN (Autonomous land vehicle in a neural network) [124], [125], [126], [127], AURORA (Automotive Run-OffRoad Avoidance) [21], and ALVINN-VC (for Virtual Camera) [61], [63], etc. A measure of the success of Navlab-based systems was the "No hands across America" test drive that consisted of a 2,849 mile trip from Pittsburgh, Pennsylvania, to San Diego, California.…”

Autonomous exploration using rapid perception of low-resolution image information

“…For example, in developing a tour-giving autonomous navigating robot, Horswill [15] used 64 × 48 images for navigation and 16 × 12 images for place recognition. Similarly, the ALVINN neural network controlled the autonomous CMU Navlab using just 30 × 32 images as input [33]. Robust obstacle avoidance was achieved by Lorigo et al [23] using 64 × 48 images.…”

Motion planning in urban environments

“…Approaches to road following in these and other systems vary drastically. For example, one of the road following algorithms used by NavLab vehicles is ALVINN (Pomerleau, 1991), which learns the mapping from road images onto control commands using a neural network by observing how the car is driven manually for several minutes. In the VITA project, on the other hand, the planner was used to track the lane while regulating the velocity of the vehicle in response to the curvature of the road and the distance to nearby vehicles and obstacles.…”