Shuqing Zeng

Dav: a humanoid robot platform for autonomous mental development

A humanoid robot, called Dav, was developed at Michigan State University as a testbed for experimental investigations into autonomous mental development. This general-purpose humanoid platform consists of a total of 43 degrees of freedom (DOF), including a drive base, torso, arms, hands, neck and head. The body may support a wide array of locomotive and manipulative behaviors. For perception, Dav is equipped with a variety of sensing systems, including visual, auditory and haptic sensors. Its computational resources are totally on-board, including quadruple Pentium III+ PowerPCs, large memory and storage, networks, and a long-sustention power supply. We discuss major design issues of a developmental humanoid and the design characteristics of the Dav robot in this paper.

Online-learning and Attention-based Approach to Obstacle Avoidance Using a Range Finder

The problem of developing local reactive obstacle-avoidance behaviors by a mobile robot through online real-time learning is considered. The robot operated in an unknown bounded 2-D environment populated by static or moving obstacles (with slow speeds) of arbitrary shape. The sensory perception was based on a laser range finder. A learning-based approach to the problem is presented. To greatly reduce the number of training samples needed, an attentional mechanism was used. An efficient, real-time implementation of the approach was tested, demonstrating smooth obstacle-avoidance behaviors in a corridor with a crowd of moving students as well as static obstacles.

Motor initiated expectation through top-down connections as abstract context in a physical world

Recently, it has been shown that top-down connections improve recognition in supervised learning. In the work presented here, we show how top-down connections represent temporal context as expectation and how such expectation assists perception in a continuously changing physical world, with which an agent interacts during its developmental learning. In experiments in object recognition and vehicle recognition using two types of networks (which derive either global or local features), it is shown how expectation greatly improves performance, to nearly 100% after the transition periods. We also analyze why expectation will improve performance in such real world contexts.

A THEORY OF DEVELOPMENTAL MENTAL ARCHITECTURE AND THE DAV ARCHITECTURE DESIGN

The software architecture of a developmental robot is a challenging new research subject. This paper presents a theory of developmental mental architecture. Five architecture types, from the simplest Type-1 (observation-driven Markov decision process) to Type-5 (DOSASE MDP), are introduced. The properties and limitations of a simpler one are discussed before the introduction of the next more complex one. Further, we present the architecture design of the Dav robot, a humanoid robot built in the Embodied Intelligence Laboratory at Michigan State University. The framework of the Dav architecture is hand-designed, but the actual controller is developed, i.e. generated autonomously by the developmental program through real-time, online interactions with the real physical environment. We present the Dav architecture and the major components that realize the architecture. The designed architecture for Dav is the next generation version from its extensively tested predecessor, the SAIL developmental robot. Closely related to the issue of performance metrics, the paper also introduces the notion of intelligence completeness (concept completeness, intelligence-metric completeness, and factor completeness) and establishes the concept of the completeness theorem for developmental robotics.

Incremental Online Object Learning in a Vehicular Radar-Vision Fusion Framework

In this paper, we propose an object learning system that incorporates sensory information from an automotive radar system and a video camera. The radar system provides coarse attention for the focus of visual analysis on relatively small areas within the image plane. The attended visual areas are coded and learned by a three-layer neural network utilizing what is called in-place learning: Each neuron is responsible for the learning of its own processing characteristics within the connected network environment, through inhibitory and excitatory connections with other neurons. The modeled bottom-up, lateral, and top-down connections in the network enable sensory sparse coding, unsupervised learning, and supervised learning to occur concurrently. This paper is applied to learn two types of encountered objects in multiple outdoor driving settings. Cross-validation results show that the overall recognition accuracy is above 95% for the radar-attended window images. In comparison with the uncoded representation and purely unsupervised learning (without top-down connection), the proposed network improves the overall recognition rate by 15.93% and 6.35%, respectively. The proposed system is also compared favorably with other learning algorithms. The result indicates that our learning system is the only one that is fit for incremental and online object learning in a real-time driving environment.

Obstacle avoidance through incremental learning with attention selection

This work presents a learning-based approach to the task of generating local reactive obstacle avoidance. The learning is performed online in real-time by a mobile robot. The robot operated in an unknown bounded 2-D environment populated by static or moving obstacles (with slow speeds) of arbitrary shape. The sensory perception was based on a laser range finder. To greatly reduce the number of training samples needed, an attentional mechanism was used. An efficient, real-time implementation of the approach had been tested, demonstrating smooth obstacle-avoidance behaviors in a corridor with a crowd of moving students as well as static obstacles.

An Object-Tracking Algorithm for 3-D Range Data Using Motion and Surface Estimation

Given a series of point sets sampled from a rigid surface by a 3-D rangefinder, we study the problem of estimating the motion and surface structure of a dynamic object. This target tracking problem with 3-D data can be formulated as maximizing the likelihood of the data (the scan map) and the Gaussian mixture model (GMM; object model up to the previous time step). We choose the prior for the object model from the conjugate distribution family of the GMM to yield a trackable posterior distribution for the object model. This GMM-based nonparametric model can be indexed by a hash lookup table, and we show that the methods complexity linearly scales with the number of scan points. Quantitative performance evaluation demonstrates that the proposed method substantially outperforms others. Results of road tests in divided freeway and urban scenes show the accuracy and robustness of the system, which can enable many vehicle active-safety and driver-assistance applications.

Online Injection of Teacher's Abstract Concepts into a Real-time Developmental Robot with Autonomous Navigation as Example

An autonomous developmental robot typically requires a considerable amount of developmental experience before it is able to learn, master and use abstract concepts, even if its speed of cognitive development is not limited by the speed of cell growth in humans. How can a developmental robot learn abstract concepts early on and use these concepts to reason and make decision? This paper introduces a frame work of two macro-layers. The upper macro-layer enables human teachers to interactively inject a representation of abstract concepts (e.g., location) into the developmental process. The lower layer takes the desired information (e.g., desired heading direction) as the input. Autonomous navigation with a global path planner is experimented with as example

Gradient sparse optimization via competitive learning

In this paper, we propose a new method to achieve sparseness via a competitive learning principle for the linear kernel regression and classification task. We form the duality of the LASSO criteria, and transfer an /spl lscr/ /sub 1/ norm minimization to an /spl lscr//sub /spl infin// norm maximization problem. We introduce a novel solution derived from gradient descending, which links the sparse representation and the competitive learning scheme. This framework is applicable to a variety of problems, such as regression, classification, feature selection, and data clustering.