In the process of building speech recognition models, accurate labeling of speech utterances is extremely time consuming and requires trained linguists. For fast building the speech recognition models in some industri...
详细信息
Path planning is one of the main sectors of robotics that deals with path calculations from a starting point to a goal in a defined field. Physical algorithms that rely on artificial potential fields and fluid fields ...
详细信息
ISBN:
(纸本)9781665483773
Path planning is one of the main sectors of robotics that deals with path calculations from a starting point to a goal in a defined field. Physical algorithms that rely on artificial potential fields and fluid fields have been developed to solve the path planning task. The main problem for the former is the creation of local minima in the field which requires extra algorithms to solve with increased path cost and added inefficiencies, while for the latter, the problem is the computational cost. This paper proposes an enhanced computational-physical path planning algorithm based on fluid stream equations and mechanics - named the Stream Field Navigation (SFN) - with modifications on how the residuals are calculated to improve the computational efficiency by introducing the Directional Residuals. SFN also introduces the Stream Reversal approach to represent a navigation field with no local minima. A comparison between the SFN algorithm and the Artificial Potential Field method is carried out to show the fundamental differences between the two methods and the results include empirical comparisons between the execution times of multiple algorithms which show that the SFN algorithm is at least 75% faster than PRM while A* is the fastest approach.
This paper proposes a calibration strategy for distorted camera with high-precision. It makes use of planar homogrphy constraint to estimate intrinsic and extrinsic parameter. On the basis of this initial guess, an op...
详细信息
ISBN:
(纸本)0780384636
This paper proposes a calibration strategy for distorted camera with high-precision. It makes use of planar homogrphy constraint to estimate intrinsic and extrinsic parameter. On the basis of this initial guess, an optimization scheme is used to minimize a new cost function, which is based on a Perspective 3-Point (P3P) algorithm. All the calibration parameters are globally optimized simultaneously with Genetic Algorithm(GA). Simulation and real image experiment result show this technique have much higher precision than traditional methods.
Relocalization is one of the necessary modules for mobile robots in long-term autonomous movement in an environment. Currently, visual relocalization algorithms mainly include feature-based methods and CNN-based (Conv...
详细信息
Deep Neural Networks, particularly Convolutional Neural Networks (ConvNets), have achieved incredible success in many vision tasks, but they usually require millions of parameters for good accuracy performance. With i...
详细信息
ISBN:
(数字)9781665490627
ISBN:
(纸本)9781665490627
Deep Neural Networks, particularly Convolutional Neural Networks (ConvNets), have achieved incredible success in many vision tasks, but they usually require millions of parameters for good accuracy performance. With increasing applications that use ConvNets, updating hundreds of networks for multiple tasks on an embedded device can be costly in terms of memory, bandwidth, and energy. Approaches to reduce this cost include model compression and parameter-efficient models that adapt a subset of network layers for each new task. This work proposes a novel parameter-efficient kernel modulation (KM) method that adapts all parameters of a base network instead of a subset of layers. KM uses lightweight task-specialized kernel modulators that require only an additional 1.4% of the base network parameters. With multiple tasks, only the task-specialized KM weights are communicated and stored on the end-user device. We applied this method in training ConvNets for Transfer Learning and Meta-Learning scenarios. Our results show that KM delivers up to 9% higher accuracy compared to other parameter-efficient methods on the Transfer Learning benchmark.
Needle puncture is irreplaceable for some percutaneous therapies, like biopsy, brachytherapy treatment, neurosurgery, etc. But needle insertion force deforms soft tissue, making accurate targeting difficult. It thus e...
详细信息
Recently, high throughput protein crystallization requires that the volumes of sample solution in crystallization experiments should be reduced to increase the screening chance with a given amount of sample. This lead...
详细信息
We discuss the mechanical bases of cellular motility by swimming and crawling. Special emphasis placed on the connections between low Reynolds number swimming and Geometric Control Theory, and on the geometric structu...
详细信息
We discuss the mechanical bases of cellular motility by swimming and crawling. Special emphasis placed on the connections between low Reynolds number swimming and Geometric Control Theory, and on the geometric structure of the underlying equations of motion. We examine some concrete examples, taken from the case studies that have been recently considered by our group. These include reverse engineering of the euglenoid movement, self-propelled droplets of active fluids, and one-dimensional models of slender crawlers. (C) 2015, IFAC (international Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved,
An adaptive fuzzy logic control scheme is developed for tracking a square trajectory by the endpoint of a two-link rigid joint space robot. The control strategy is based on a direct adaptive control scheme in which th...
详细信息
An adaptive fuzzy logic control scheme is developed for tracking a square trajectory by the endpoint of a two-link rigid joint space robot. The control strategy is based on a direct adaptive control scheme in which the controller gains are adapted in real-time according to fuzzy logic systems such that the tracking errors between a reference model and the actual robot system outputs are brought to zero.
Machining using industrial robots is currently limited to applications with low geometrical accuracies and soft materials due to weaknesses of the robot structure, insufficient controller performance and the lack of s...
详细信息
ISBN:
(纸本)9783642392221
Machining using industrial robots is currently limited to applications with low geometrical accuracies and soft materials due to weaknesses of the robot structure, insufficient controller performance and the lack of suitable software tools. This paper proposes a modular approach to overcome these obstacles, applied both during program generation (offline) and execution (online). Offline predictive machining errors compensation is achieved by means of an innovative programming system, based on kinematic and dynamic robot models. Realtime adaptive machining error compensation is also provided by sensing the real robot positions with an innovative tracking system and corrective feedback to both the robot and an additional high dynamic compensation mechanism on piezo-actuator basis. Due to the modularity of the approach, an individual setup can be compiled for each actual use-case. Final experimental validation of the components is currently ongoing in multiple robot cells, covering several application areas as aerospace, automotive or mould construction.
暂无评论