Neural Network Perception for Mobile Robot Guidance

Dean Pomerleau's trainable road tracker, ALVINN, is arguably the world's most famous neural net application. It currently holds the world's record for distance traveled by an autonomous robot without interruption: 21.2 miles along a highway, in traffic, at speedsofup to 55 miles per hour. Pomerleau's work has received worldwide attention, including articles in Business Week (March 2, 1992), Discover (July, 1992), and German and Japanese science magazines. It has been featured in two PBS series, "The Machine That Changed the World" and "By the Year 2000," and appeared in news segments on CNN, the Canadian news and entertainment program "Live It Up", and the Danish science program "Chaos". What makes ALVINN especially appealing is that it does not merely drive - it learns to drive, by watching a human driver for roughly five minutes. The training inputstothe neural networkare a video imageoftheroad ahead and thecurrentposition of the steering wheel. ALVINN has learned to drive on single lane, multi-lane, and unpaved roads. It rapidly adapts to other sensors: it learned to drive at night using laser reflectance imaging, and by using a laser rangefinder it learned to swerve to avoid obstacles and maintain a fixed distance from a row of parked cars. It has even learned to drive backwards.

Springer Book Archives

1 Introduction.- 1.1 Problem Description.- 1.2 Robot Testbed Description.- 1.3 Overview.- 2 Network Architecture.- 2.1 Architecture Overview.- 2.2 Input Representations.- 2.3 Output Representation.- 2.4 Internal Network Structures.- 3 Training Networks "On-The-Fly".- 3.1 Training with Simulated Data.- 3.2 Training "on-the-fly" with Real Data.- 3.3 Performance Improvement Using Transformations.- 3.4 Discussion.- 4 Training Networks With Structured Noise.- 4.1 Transitory Feature Problem.- 4.2 Training with Gaussian Noise.- 4.3 Characteristics of Structured Noise.- 4.4 Training with Structured Noise.- 4.5 Improvement from Structured Noise Training.- 4.6 Discussion.- 5 Driving Results and Performance.- 5.1 Situations Encountered.- 5.2 Driving with Alternative Sensors.- 5.3 Quantitative Performance Analysis.- 5.4 Discussion.- 6 Analysis of Network Representations.- 6.1 Weight Diagram Interpretation.- 6.2 Sensitivity Analysis.- 6.3 Discussion.- 7 Rule-Based Multi-network Arbitration.- 7.1 Symbolic Knowledge and Reasoning.- 7.2 Rule-based Driving Module Integration.- 7.3 Analysis and Discussion.- 8 Output Appearance Reliability Estimation.- 8.1 Review of Previous Arbitration Techniques.- 8.2 OARE Details.- 8.3 Results Using OARE.- 8.4 Shortcomings of OARE.- 9 Input Reconstruction Reliability Estimation.- 9.1 The IRRE Idea.- 9.2 Network Inversion.- 9.3 Backdriving the Hidden Units.- 9.4 Autoencoding the Input.- 9.5 Discussion.- 10 Other Applications - The SM2.- 10.1 The Task.- 10.2 Network Architecture.- 10.3 Network Training and Performance.- 10.4 Discussion.- 11 Other Vision-based Robot Guidance Methods.- 11.1 Non-learning Autonomous Driving Systems.- 11.2 Other Connectionist Navigation Systems.- 11.3 Other Potential Connectionist Methods.- 11.4 Other MachineLearning Techniques.- 11.5 Discussion.- 12 Conclusion.- 12.1 Contributions.- 12.2 Future Work.