Smart wheelchair navigation system: current status and future directions

0 introduction

Smart wheelchair areas as medical care, and its application service robot makes extensive use of a mobile robot technology.

In the smart wheelchair research involves key technologies have navigation system, control, and energy systems, human-machine interface, but since the entire wheelchair system to people-centered, so in research to solve is the core of the issue of safe navigation wheel. The so-called navigation that refers to the mobile robot according to the advance given task command, according to the known map information to make a global path planning and travel process, constantly perception surrounding local environmental information, make decisions independently, and adjust their position at any time, boot itself safe driving arrives at the destination location.

This article on intelligent wheelchair navigation of core issues analysis, points out the current research involves the technical progress and shortcomings, and its development trends provides a brief description.

1 system orientation

Smart wheelchair positioning is the environment information is in motion, use of their own sensors to determine its real time in the work environment reference coordinates relative to the global coordinate of the position and posture.

Location technology can be divided into 2 main categories: machine vision of positioning technologies and non-computer based sensor positioning technology.

Common positioning methods have Optical Encoder, inertial gyro and magnetic compasses, road signs etc. Each method has strengths and limitations, in practice smart wheelchair actual integrated employs several methods to improve the positioning accuracy and reliability of the system, but accuracy from users the normal use of a certain distance, how to improve the positioning precision and effectiveness in the future to focus on one of the issues.

Sensor selection in positioning is very important.

According to location technology, the sensor can be divided into Visual and non-vision sensors. There are currently popular sensors ultrasonic distance measuring sensors, CCD cameras, infrared sensors, laser sensor, GPS, etc. Because the ultrasonic obstacle avoidance for convenience, the technology mature, low cost, become smart wheelchair usual location method, the application uses multiple ultrasonic distance sensors, ultrasonic distance measuring sensor detection distance obstructions, and then determine the robot is currently located.

2 information fusion

Information received from the sensors do not guarantee completely reliable and correctly, may cause the physical existence of aliasing or distance detection error, then we can use probability method, integrated numerous observations, multi-sensor fusion methods for processing, multi-sensor fusion method to become a hotspot in recent years.

The so-called information fusion can be generalized to overview for such a process, that data from multiple sensors and information according to established rules analysis, combining into one comprehensive intelligence reports, and on this basis, the demand for system users provide information such as: decisions, track tasks, and so on.

In sensor data fusion, using multiple types of sensor is necessary. Multisensor data fusion technology has demonstrated single sensor incomparable advantages, synthesizer, you can get from any single input data get more reliable information.

How to blend these complementary or redundant sensor information and be more fully reflect the environmental characteristics of information method is particularly important.

In the study of the most crucial part is the information fusion algorithm, it has proposed a variety of applications in different systems of multi-sensor data fusion, these algorithms can be divided into 2 categories: Random class methods and the methods of artificial intelligence.

(1) random class methods

This class of object is random, in multi-sensor fusion often uses the random class methods including many, such as: the weighted average method, statistical decision theory, cluster analysis, wavelet transform, Bayes reasoning method, Dempster-Sharer evidence theory, Kalman filtering fusion, etc.

(2) artificial intelligence methods

In recent years for multi-sensor data fusion of computational intelligence methods are: fuzzy set theory, expert systems, neural networks, rough set and support vector machine, etc.

Which neural network based on a variety of sensor data fusion is recent years development of the HotSpot. Neural network with good tolerance, stratification, plasticity, adaptability, associative memory and parallel processing capability, neural networks and other methods of combining information fusion research, the results are obvious, others formed a kind of trend. For example, wavelet and neural networks, Kalman filtering and neural networks, Dempster-Shafer evidence theory and neural networks, fuzzy and neural networks, genetic algorithms and neural networks.

Future of multi-sensor data fusion technology mainly concentrated in the algorithm and the emergence of the new algorithm, the development of micro-sensors, as well as multi-level information fusion 3.

3 path planning issues

Route planning is an obstacle, as smart wheelchair from starting point to end point looking for a no-touch path, and in accordance with certain principles for optimization, find a shortest path.

Path planning contains two aspects: the first is the environmental model; second, path planning algorithm design.

(1) environmental modeling

Path planning for environmental modeling is a precondition for statically known environment, many successful research results, its also more sophisticated modeling techniques.

For some known or totally unknown environment path planning has never been the perfect solution, the root cause lies in the resolution on the environment and the amount of environmental information storage.

Environmental modeling in general there are three types: network Pier figure model, grid model and hierarchical model.

Network and conpoy model includes free space law, vertex image method, generalized cone method, is on the environment of high level description, computing a large sensor accuracy higher; grid model is the same size as the space is divided into a grid, modeling simple, but the search space very big; levelStructure of the model is based on the data range criterion of consistency and recursively decompose principle on the space environment for modeling, use this kind of model compression the search space, and is very easy to use sensor information to update the model.

(2) the path planning algorithm

Under smart wheelchair on environmental information about different, path planning can be divided into 2 types: global path planning and local path planning.

Where global path planning need to be aware of all the messages on the environment, and produce a series of critical points as a destination point release to the local path planning system. The local path planning you only need to distance robot more recent information, the obstacles in motion sensor information to continually update its internal environment information, planning out a line from the starting point or a child of the target point to the next child of the target point of the path. Comparison of some path search algorithm to find better solutions; further research on the activities of the trend analysis, obstacles to the obstacle avoidance strategy this 2 aspects is the robot path planning to address key issues. According to the environmental information to understand the degree of integrity, path planning may use different algorithms. For global path planning often used algorithms are: can view, free-space method and grid method and so on. Local path planning common methods are: the artificial potential field, genetic algorithm and fuzzy logic algorithms, etc.

In recent years in these traditional methods, these methods have been further integration and extended, such as route planning based on genetic algorithm--simplified coding of two-dimensional path is a one-dimensional path encoding problems, fuzzy neural network obstacle avoidance method--based on the actual error function and membership function method, based on laser radar path planning methods--point of potential field, with virtual force field--dynamic grid method and potential field combination.

4 conclusions

Above for smart wheelchair navigation research methods and ideas of a more comprehensive discussion, at the same time these methods also apply to mobile robots.

With good planning capacity, real-time and practical research on intelligent wheelchair, is the core technology for future research, and is also the focus in the area of robotics research and difficult problems.