Hardware realization of neural network based controller for autonomous robot navigation

Aamer, Najmuddin; Ramachandran, S.

doi:10.1109/iccmc.2017.8282684

Cited by 3 publications

(2 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For instance, AlexNet, a convolutional neural network with 8 layers, has 60 million network parameters, which require 240MB of memory for storage as 32-bit numbers [32]. These types of implementation are popularly realized in field programmable gate arrays (FPGA) [33,34] because of the flexibility to change the neural weights for different applications. For small neural networks, FPGA implementations are faster [35] than executing software programs on microprocessors.…”

Section: Implementing Neural Network Circuitsmentioning

confidence: 99%

Recurrent neural network implementation of digital integrated circuits to mitigate challenges in design verification

Yeong,

2023

View full text Add to dashboard Cite

Design verification is the dominant stage that consumes the most resources in the digital integrated circuit (IC) design process. Design verification is important because human designers imperfectly convert high-level specifications to low-level circuit implementations using standard cell logic, which is nonlinear and complex to predict and characterize. The widening process variations in shrinking process technologies while digital designs grow in scale and complexity to the extent of being impossible to fully or intuitively identify all temporal interactions of a specific design. Deep neural networks (DNN) are being progressively integrated into the sophisticated software tool chain and design process flow of digital IC as artificial intelligence can learn relationships and correlations in complex, high-dimensional, and multi-factorial problems. In this work, we propose to apply DNN to implement digital IC to minimize the complexity of digital design verification. We posit that DNN can learn to implement circuit functions directly from high-level specifications without requiring detailed specifications from the designer. Trained neural networks can be implemented on neuromorphic hardware to achieve greater power and compute efficiencies than the conventional standard cell implementation. We demonstrate that over 150 randomly generated finite state machines (FSM) can be learned effectively with Recurrent Neural Network (RNN) comprising Gated Recurrent Units (GRU) with different complexity as indicated by the number of states and inputs to the FSM. Our proposed methodology of learning RNN GRU implementations of FSM demonstrates a way forward to reduce the cost and effort of design verification, ultimately leading towards faster digital IC design cycles.

show abstract

Section: Implementing Neural Network Circuitsmentioning

confidence: 99%

Recurrent neural network implementation of digital integrated circuits to mitigate challenges in design verification

Yeong,

2023

View full text Add to dashboard Cite

show abstract

“…The path planning of a mobile robot usually means that a mobile robot can find an optimal path or near optimal path [1] from the starting point to avoid the obstacle arriving at the target point from the starting point in the obstacle environment. At present, the main approaches to solve the path planning problems include artificial potential field [2] , immune algorithm [3] , genetic algorithm [4] , artificial neural network [5] and so on. But these algorithms are complex, locally optimal and inefficient.…”

Section: Introductionmentioning

confidence: 99%

Path Planning Based on an Improved Ant Colony Algorithm

Wang

Zhao

et al. 2018

MATEC Web Conf.

View full text Add to dashboard Cite

The classical ant colony algorithm has the disadvantages of initial search blindness, slow convergence speed and easy to fall into local optimum when applied to mobile robot path planning. This paper presents an improved ant colony algorithm in order to solve these disadvantages. First, the algorithm use A* search algorithm for initial search to generate uneven initial pheromone distribution to solve the initial search blindness problem. At the same time, the algorithm also limits the pheromone concentration to avoid local optimum. Then, the algorithm optimizes the transfer probability and adopts the pheromone update rule of "incentive and suppression strategy" to accelerate the convergence speed. Finally, the algorithm builds an adaptive model of pheromone coefficient to make the pheromone coefficient adjustment self-adaptive to avoid falling into a local minimum. The results proved that the proposed algorithm is practical and effective.

show abstract

Object Recognition Development for Android Mobile Devices with Text-to-Speech Function Created for Visually Impaired People

Burta

Szabó

Gontean

2020

2020 Fourth World Conference on Smart Trends in Systems, Security and Sustainability (WorldS4)

View full text Add to dashboard Cite

Hardware realization of neural network based controller for autonomous robot navigation

Cited by 3 publications

References 14 publications

Recurrent neural network implementation of digital integrated circuits to mitigate challenges in design verification

Recurrent neural network implementation of digital integrated circuits to mitigate challenges in design verification

Path Planning Based on an Improved Ant Colony Algorithm

Object Recognition Development for Android Mobile Devices with Text-to-Speech Function Created for Visually Impaired People

Contact Info

Product

Resources

About