We provide here a novel method, called hypercolumn sparsification, to achieve high recognition performance for convolutional neural networks (CNNs) despite low-precision weights and activities during both training and test phases. This method is applicable to any CNN architecture that operates on signal patterns (e.g., audio, image, video) to extract information such as class membership. It operates on the stack of feature maps in each of the cascading feature matching and pooling layers through the processing hierarchy of the CNN by an explicit competitive process ( k -WTA, winner take all) that generates a sparse feature vector at each spatial location. This principle is inspired by local brain circuits, where neurons tuned to respond to different patterns in the incoming signals from an upstream region inhibit each other using interneurons, such that only the ones that are maximally activated survive the quenching threshold. We show this process of sparsification is critical for probabilistic learning of low-precision weights and bias terms, thereby making pattern recognition amenable for energy-efficient hardware implementations. Further, we show that hypercolumn sparsification could lead to more data-efficient learning as well as having an emergent property of significantly pruning down the number of connections in the network. A theoretical account and empirical analysis are provided to understand these effects better.

A set of architectural concepts that address the needs for integrating high-level planning activities with lower-level reactive or participatory behaviors is presented. Based on lessons learned from the authors' experience with a hierarchical architecture for autonomous cross-country navigation, various pitfalls that may arise from the misuse of abstraction have been recognized. Consequently, a new approach that emphasizes the minimization of information loss both within and between system layers has been adopted. This change in perspective has allowed the authors to greatly enhance the overall capabilities and performance of their system.>

A description is given of the first cross-country map and sensor-based autonomous operation of a robotic vehicle. Experiments on the autonomous land vehicle (ALV) in natural terrain were performed. An overview of the software architecture is provided, and details of the perception and planning techniques are presented. Two key experiments in which the vehicle avoided known and unknown obstacles in its path are described.>

We describe a software architecture to support the planning and control requirements of an autonomous land vehicle. This architecture is designed specifically to handle diverse terrain with maximal speed, efficacy and versatility through the use of a library of reflexive strategies specialized to particular needs. A hierarchy of control is built in which lower level modules perform tasks requiring greatest immediacy while higher level modules perform tasks involving greater assimilation of sensor data. With all levels of the hierarchy operating in parallel, higher level modules perform selective activation and deactivation of lower level responses with limited interference from demands for immediacy, allowing the system to exploit the advantages of both immediate and assimilated data. The reflexive component of this system has been demonstrated both in a detailed real-time simulation and on a small indoor robotic vehicle.