The ADAPT project is a collaboration of three universities building a unified architecture for mobile robots. The goal of this project is to endow robots with the full range of cognitive abilities, including perceptio...
详细信息
ISBN:
(纸本)9780769532721
The ADAPT project is a collaboration of three universities building a unified architecture for mobile robots. The goal of this project is to endow robots with the full range of cognitive abilities, including perception, use of natural language, learning and the ability to solve complex problems. The perspective of this work is that such an architecture should be based on language and visualization. ADAPT is based on an the structure and semantics of language, and more specifically on algebraic linguistics and visualization of semantics. ADAPT organizes its knowledge using linguistic robot schemas, which implement linguistic units within a concurrent, distributed programming language. Each schema is associated with one or more 3D visualizations that provide its semantics. These visualizations are dynamic, and are composed within a virtual world to create ADAPTpsilas representation of itself and its environment.
Autonomous vehicles can significantly improve efficiency and safety in applications ranging from warfare to transportation. However, to supply those benefits they must be shown to operate effectively, safely, and reli...
详细信息
ISBN:
(纸本)9780769532721
Autonomous vehicles can significantly improve efficiency and safety in applications ranging from warfare to transportation. However, to supply those benefits they must be shown to operate effectively, safely, and reliably in a wide range of terrains and conditions. Most major successes with autonomous vehicles have been limited to somewhat structured environments. We are interested in autonomous vehicles that can operate in forested areas, which are one of the most unstructured and difficult terrains due to the high number and varied nature of potential obstacles, the complexity of the visual field, and the difficulty in getting a good GPS fix due to overhead interference. In this paper we present a novel control system designed with the eventual goal of forest operation. It is built on top of the learning applied to ground robotics (LAGR) system developed at Carnegie Mellon. The new control system consists of the University of Idaho (UI) LAGR Planner and the UI software for LAGR vision system. The results show that the combination of these two modules significantly improve the capabilities of the LAGR robot and, more importantly, allow it to perform autonomously in complex environments such as primitive forest trails that the base system could not navigate.
Accurate models of the robot dynamics allow the design of significantly more precise, energy-efficient and more compliant computed torque control for robots. However, in some cases the accuracy of rigid-body models do...
详细信息
ISBN:
(纸本)9780769532721
Accurate models of the robot dynamics allow the design of significantly more precise, energy-efficient and more compliant computed torque control for robots. However, in some cases the accuracy of rigid-body models does not suffice for sound control performance due to unmodeled nonlinearities such as hydraulic cables, complex friction, or actuator dynamics. In such cases, learning the models from data poses an interesting alternative and estimating the dynamics model using regression techniques becomes an important problem. However, the most accurate regression methods, e.g. Gaussian processes regression (GPR) and support vector regression (SVR), suffer from exceptional high computational complexity which prevents their usage for large numbers of samples or online learning to date. We proposed an approximation to the standard GPR using local Gaussian processes models inspired by [Vijayakumaf, 2005 and Snelson and Ghahramani, 2007]. Due to reduced computational cost, local Gaussian processes (LGP) is capable for an online learning. Comparisons with other nonparametric regressions, e.g. standard GPR, v-SVR and locally weighted projection regression (LWPR), show that LGP has higher accuracy than LWPR and close to the performance of standard GPR and v-SVR while being sufficiently fast for online learning.
While most research in social robotics embraces the challenge of designing and studying the interaction between robots and humans itself, this talk will discuss the utility of social interaction in order to facilitate...
While most research in social robotics embraces the challenge of designing and studying the interaction between robots and humans itself, this talk will discuss the utility of social interaction in order to facilitate for more flexible robotics. What can a robot gain with respect to learning and adaptation from being able to sociably interact? What are basic learning-enabling behaviors? And how do inexperienced human tutor robots a sociable way? In order to answer these question we consider the challenge of learning by interaction as a systemic one, comprising appropriate perception, system design, and feedback. Basic abilities of robots will be outlined which resemble concepts of developmental learning in infants, apply linguistic models of interaction management, and take tutoring as a joint task of a human and a robot. However, in order to tackle the challenge of learning by interaction the robot has to couple and coordinate these behaviors in a very flexible and adaptive manner. The active memory as an architectural concept in particular suitable for learning-enabled robots will be briefly discussed as a foundation for coordination and integration of such interactive roboticsystems. The talk will build a bridge from the construction of integrated roboticsystems to their evaluation, analysis, and way back. It will outline why we intend to enable our robots to learn by interacting and how this paradigm impacts the design of systems and interaction behaviors.
This paper presents a software implementation of a user adaptive fuzzy control system for autonomous navigation in mobile robots for unknown environments. This system has been tested in a pioneer mobile robot and on a...
详细信息
ISBN:
(纸本)1424408296;97
This paper presents a software implementation of a user adaptive fuzzy control system for autonomous navigation in mobile robots for unknown environments. This system has been tested in a pioneer mobile robot and on a robotic wheelchair, fitted with PLS laser sensor to detect the obstacles and odometry sensors for localization of robots and the goal positions. The system is able to drive the robots to their goal position avoiding static and dynamic obstacles, without using any pre-built map. Our approach learn from user behaviors in the way it can resolve different situations against obstacles or *** propose and implement two updates for the fuzzy system. For the implementation of the learning algorithm we use a weighting scheme giving a value for each fuzzy-rule, this value is based on the synapse-weight idea and represent the contribution of each rule in the system output. We also create of a more important sector in the definition of the fuzzy-variables, based on a statistics system that measure the uses of all the sets of the variables in order to contract the size of the rule-base.
We present a review of recent trends in cognitive robotics that deal with online learning approaches to the acquisition of knowledge, control strategies and behaviors of a cognitive robot or agent. Along this line we ...
详细信息
It is important for a robot to acquire adaptivebehaviors for avoiding surrounding robots and obstacles in complicated environments. Although the introduction of a learning scheme is expected to be one of the solution...
详细信息
It is important for a robot to acquire adaptivebehaviors for avoiding surrounding robots and obstacles in complicated environments. Although the introduction of a learning scheme is expected to be one of the solutions for this purpose, a large size of memory and a large calculation cost are required to handle useful information such as motions of robots. In this paper, we introduce the multi-layered reinforcement learning method. By dividing a learning curriculum into multiple layers, the number of expected situations can be reduced. It is shown that real robots can adaptively avoid collision with each other and to obstacles in a complicated situation. (C) 1999 Elsevier Science B.V. All right reserved.
暂无评论