This paper presents a novel motion planning and tracking framework for automated vehicles based on artificial potential field (APF) elaborated resistance approach. Motion planning is one of the key parts of autonomous driving, which plans a sequence of movement states to help vehicles drive safely, comfortably, economically, human-like, etc. In this paper, the APF method is used to assign different potential functions to different obstacles and road boundaries; while the drivable area is meshed and assigned resistance values in each edge based on the potential functions. A local current comparison method is employed to find a collision-free path. As opposed to a path, the vehicle motion or trajectory should be planned spatiotemporally. Therefore, the entire planning process is divided into two spaces, namely the virtual and actual. In the virtual space, the vehicle trajectory is predicted and executed step by step over a short horizon with the current vehicle speed. Then, the predicted trajectory is evaluated to decide if the speed should be kept or changed. Finally, it will be sent to the actual space, where an experimentally validated Carsim model controlled by a model predictive controller is used to track the planned trajectory. Several case studies are presented to demonstrate the effectiveness of the proposed framework.

A Motion Planning and Tracking Framework for Autonomous Vehicles Based on Artificial Potential Field Elaborated Resistance Network Approach

The feasibility of using neural networks in a practical halftoning application is considered. The cellular neural network (CNN) architecture is chosen for its proven implementability in VLSI and high-speed operation. Since both the CNN and halftoning have a geometrically local character, the CNN provides a natural implementation. The CNN template weights are derived by analogy to the well-known error diffusion algorithm for halftoning. Some limitations of the neural network approach are analyzed, providing an advance in designing template weights over previous methods. These limitations are shown to be especially critical in the case of the small interconnection neighbourhoods needed for efficient implementation. The design criteria are validated by direct simulation. The resulting halftones are shown to be more faithful reproductions of the original than those produced by the error diffusion algorithm. It is suggested that a CNN with optical inputs could provide a high-speed scanner/halftoner for applications such as facsimile. >

Image halftoning with cellular neural networks

This paper presents a novel local motion planning framework in a hierarchical manner for autonomous vehicles to follow a trajectory and agilely avoid obstacles. In the upper layer, a new path-planning method based on the resistance network is applied to plan behaviors (e.g. lane keeping or changing), where the human-like factors can be included to simulate different driver styles, such as the aggressive, moderate, and conservative. The planned results (i.e. the lane-change command and the local planned path) will guide the lower-layer planner to decide the local motion. In the lower layer, for the sake of simplicity and alleviation of the computational burden, two separate model predictive controllers (MPC) based on a point-mass kinematic model are utilized for both longitudinal and lateral motion planning. Finally, a super-twisting sliding mode controller (STSMC) based motion tracker is designed to show the feasibility of the proposed decoupled planning method and decide the desired control actions of autonomous vehicles. Several scenarios are defined to comprehensively test and demonstrate the effectiveness and the real-time applicability of the new motion-planning framework. The results show that the proposed method performs very well in the planning and tracking process and takes less than $\text{25 ms}$ for the whole planning process, which can be easily implemented in real-world applications.

A Novel Local Motion Planning Framework for Autonomous Vehicles Based on Resistance Network and Model Predictive Control

Motion planning, or goal-oriented, context-sensitive, intelligent control is essential if an agent is to act in a useful manner. This paper suggests a new class of motion planners that can mark a constrained trajectory to a target zone in an environment that need not necessarily be a priori known. The novelty of the suggested planner lies in its ability to enforce region avoidance and direction satisfaction constraints jointly. To the best of the authors' knowledge, this is the first time that directional constraints have been addressed in the motion planning literature. To build such a planner, the potential field approach is used for inducing the control action. In addition, to cope with the presence of the above constraints (in particular, the directional constraints), a new type of potential field, called the nonlinear anisotropic harmonic potential field, is suggested. The planner has applications in traffic management and operations research among others. Development of the approach, proofs of correctness, and simulation results are supplied.

Motion planning in the presence of directional and regional avoidance constraints using nonlinear, anisotropic, harmonic potential fields: a physical metaphor

Parallelization of cellular neural networks on GPU

A novel constructive parallel path searching method is proposed. This method is based on selecting local maximum current in unity resistive mesh (URM) whose principle is to obtain the field expressed with the Poisson equation. A bottleneck detecting approach based on the proposed method is presented. The proposed method has been applied to the problems of LSI routing, communication path routing of a processor array, and path searching for a robot. Experimental results showed the efficiency of the proposed method. >

A resistive mesh analysis method for parallel path searching

Image compression and regeneration by nonlinear associative cellular neural networks (CNNs) that can be used as retina chips are addressed. One is a sparse Hopfield-type neural network that is called an H-type CNN and the other is its dual network and is called a DH-type CNN. Their input information sequences are given by nodes and links as voltages and currents, respectively. Their error correcting capacity (minimum basins of attraction) is decided by the minimum numbers of links of cutset and loop, respectively. Simulation results showing the performance of both types of network are reported. >

http://ieeexplore.ieee.org/iel2/640/5935/00229937.pdf

Duality theory of cellular neural networks for image compression and regeneration

Pipelining Gauss Seidel Method for Analysis of Discrete Time Cellular Neural Networks

Discrete Time Cellular Neural Networks with Two Types of Neuron Circuits for Image Coding and Their VLSI Implementations

The authors describe a CMOS digital retina chip with neurons having multi-bit output for image coding from analog to digital formats. The neuron having a 1-bit output model can easily be implemented by the use of CMOS inverters. It has individual digital inputs, a common quantized output and a comparator. The weight vector is given as a fixed numerical value by a channel conductance for pull-up PMOS and pull-down NMOS transistors, respectively. The delayed binary outputs of neurons in the neighborhood are directly connected to the digital inputs. The next state of the network is computed from the current state at some neurons in any time interval. To reduce the error between the input analog gray image and the output digital image, a two-bit image coding method is proposed. >

M. Ikegami

Papers

A resistive mesh analysis method for parallel path searching

Duality theory of cellular neural networks for image compression and regeneration

Pipelining Gauss Seidel Method for Analysis of Discrete Time Cellular Neural Networks

Discrete Time Cellular Neural Networks with Two Types of Neuron Circuits for Image Coding and Their VLSI Implementations

CMOS digital retina chip with multi-bit neurons for image coding