Scenarios requiring humans to choose from multiple seemingly optimal actions are commonplace, however standard imitation learning often fails to capture this behavior. Instead, an over-reliance on replicating expert a...
详细信息
Scenarios requiring humans to choose from multiple seemingly optimal actions are commonplace, however standard imitation learning often fails to capture this behavior. Instead, an over-reliance on replicating expert actions induces inflexible and unstable policies, leading to poor generalizability in an application. To address the problem, this paper presents the first imitation learning framework that incorporates Bayesian variational inference for learning flexible nonparametric multi-action policies, while simultaneously robustifying the policies against sources of error, by introducing and optimizing disturbances to create a richer demonstration dataset. This combinatorial approach forces the policy to adapt to challenging situations, enabling stable multi-action policies to be learned efficiently. The effectiveness of our proposed method is evaluated through simulations and real-robot experiments for a table-sweep task using the UR3 6-DOF robotic arm. Results show that, through improved flexibility and robustness, the learning performance and control safety are better than comparison methods.
The notion of ThisType has been proposed to promote typesafe reuse of binary methods and recently extended to mutually recursive definitions. It is well-known, however, that ThisType does not match with subtyping well...
详细信息
ISBN:
(纸本)9781605581668
The notion of ThisType has been proposed to promote typesafe reuse of binary methods and recently extended to mutually recursive definitions. It is well-known, however, that ThisType does not match with subtyping well. In the current type systems, type safety is guaranteed by the sacrifice of subtyping, hence dynamic dispatch. In this paper, we propose two mechanisms, namely, nonheritable methods and local exactization to remedy the mismatch between ThisType and subtyping. We rigorously prove their safety by modeling them in a small calculus. Copyright 2009 ACM.
In recent years the human iris has established itself as a robust, highly accurate biometric. Academic research of iris biometrics;is expanding and several iris recognition products have become commercially available....
详细信息
ISBN:
(纸本)9781424419777
In recent years the human iris has established itself as a robust, highly accurate biometric. Academic research of iris biometrics;is expanding and several iris recognition products have become commercially available. Conventional iris recognition systems impose constraints with regard to a subjects proximity to the device as well as their movement. Recent innovations in iris acquisition systems and recognition algorithms have aimed to relax these constraints. This paper discusses the design of a novel prototype iris recognition system for long-range human identification. Eagle-Eyes (TM) is a multi-biometric system that is capable of acquiring, a face and two iris images from multiple humans present anywhere within its large capture volume. Eagle-Eyes uses multiple cameras with hierarchically-ordered field of views, a highly precise pan-tilt unit and a long focal length zoom lens. This solution performs enrollment, acquisition and processing of iris biometrics using a device and data agnostic architecture. Preliminary experimental results are reported.
Differential Evolution (DE) has been successfully applied to a variety of optimization problems. The performance of DE is affected by two algorithm parameters of the scaling factor and the crossover rate. Much researc...
详细信息
ISBN:
(纸本)9781728183923
Differential Evolution (DE) has been successfully applied to a variety of optimization problems. The performance of DE is affected by two algorithm parameters of the scaling factor and the crossover rate. Much research has been done in order to adaptively control the parameters. One of the most successful studies on adaptive parameter control is JADE, where the two parameter values are generated according to two probability distributions which are tuned by the parameter values in success cases. In this paper, we propose a new method that utilizes not only success information but also failure information. A measure, which indicates how much the pair of generated parameter values will lead to failure, is defined. The pair is rejected probabilistically according to the measure and a new pair is regenerated until a good pair is generated. The effect of the proposed method is shown by solving various benchmark problems.
暂无评论