Muscle electric Control dexterous hand (I.)

First, a definition of EMG control: The extraction of EMG signals can be applied to outside control technology

In biology, the action potential causes muscle contraction, and using EMG to control dexterous hand should belong to the category of Bionics.

The surface EMG signal is the sum of the action potential of the skin under which the electrodes are exposed (through sebum, solution conduction).

From the point of view of pattern recognition, the characteristic space of EMG signal can be infinitely dimensional, so theoretically, for limited mode (static gesture, dynamic gesture, coherent operation action and action intention ...). ), as long as the sample data is sufficient, the algorithm is appropriate, can be identified or separated.

So we can use EMG signals to determine the multi-objective such as shrinkage mode, strength (bottom) and gripping action.

Challenge:

1. Weak EMG signal

2. Randomness

3. Low signal-to-noise ratio

Dexterous hand represents: STANFORD/JPL utah/mit NASA Robonaut DLR Hand

Features: Multiple degrees of freedom, multi-sensor, high integration and miniaturization, under-drive

Drive mode: Tendon drive, connecting rod, liquid driver, connecting rod cross tendon, coupling linkage

Under-drive system: The system's independent control variable number is less than the number of system degrees of freedom of a class of nonlinear system, in saving energy, reduce cost, reduce weight, enhance system flexibility and other aspects are superior to the complete drive system, simply said that the input than to control the amount of less system.

The structure of the under-drive system is simple and convenient for the whole dynamic analysis and experiment. Due to the high nonlinearity of the system, parametric perturbation, multi-objective control requirements and limited control, the under-driven system is complex enough to study and verify the effectiveness of various algorithms. When the drive fails, it may make the complete drive system become under-drive system, the under-drive control algorithm can play the role of fault-tolerant control. From the point of view of control theory, the limitation of control input of under-drive system is a challenging control problem, and the research on the control problem of under-driven mechanical system is helpful to the development of nonholonomic constrained system control theory. Bridge cranes, Pendubot (pendulum Robot), Acrobot (Acrobat Robot), inverted pendulum systems are typical under-drive systems.

A review of EMG control methods

Principle:

The central nervous system controls the size and speed of muscle contraction by raising the Movement unit (motor Units,mus) (the process of moving units from small to large) and releasing rates (the number of times the unit is excited in units of time).

EMG is a one-dimensional time series signal, which originates from the motor neuron in the central nervous cord, and is the sum of the action potentials (action Potentials,aps) released by many movement units exposed to the electrodes.

As a result, EMG signals contain muscle contraction patterns and strength information that can be used as a source of control information. (This is a feedforward control, how to consider the introduction of feedback control, force information and force feedback)

Development history:

1. Threshold-based decision 1960~1970

Using the two-state amplitude and envelope demodulation method of EMG signal: After correcting, filtering and modulating the EMG signal, the primary contraction activity of the muscle is corresponding to the peak of the signals, and the hand grasping or stretching action is obtained by comparing the peak value with the threshold size.

Key Technologies: Amplitude modulation

2. Based on amplitude coding

The three-state mode of EMG signal is encoded and output according to the timing. The specific EMG control language consists of three-bit code, (A1 (T), A2 (T), A3 (t)) the AI (t) belongs to {0,1,2} each of the three states "1" "2" "0" that can use the amplitude of the EMG signal ("1" and "2" by the size of the EMG contraction amplitude, "0" Represents a non-muscular activity) they are combined to match the grasping mode and gripping force of various prosthetic hand.

Each muscle contraction test relies on a non-Gaussian evaluation of the EMG (inclination, kurtosis) and then filters within a specific time window (to remove the interfering noise) to determine whether the signal amplitude peak intensity is "1" or "2". After all three bits of the encoding language have been collected, the prosthetic hand is made compulsory (should be the mapping table), and its position/force control adopts a bottom-level compliance control method.

3. Hierarchical control decisions

By the advanced signal modulation and pattern recognition algorithm, the hand grasping mode of the muscle stump is identified, and the bottom controller realizes the stable grasping of the prosthetic prosthesis.

Example:

A. Germany adopts a method of hand control based on the combination of conversion signal and control signal, the EMG mode information is obtained by processing the conversion signal, and the control information of each finger is obtained by processing the control signal (proportional control of speed/force).

B. The University of Southampton has a similar structure for the control of the Southampton hand: the user uses the ordinary two-state control mode to achieve the grasping of the object, and adopts the microprocessor and the sensing system to obtain the feedback information of the motherland, so as to realize the self-discipline and the allocation of grasping force. "Dexterous Hand Adaptive Operation Flow" is: pre-set, contact, hold, squeeze and relax 5 states, transitions between states rely on the contraction and stretching of a single muscle to achieve

C. The time domain characteristics (over 0 pips, average absolute value, etc.) of EMG signals in Canada using different modes of operation, using artificial neural network multilayer perceptron as classifier to identify different muscle contraction patterns in 4. But its disadvantage is that the grasping mode and motion control of prosthetic hand still need the visual feedback of human eyes.

Hybrid control Method: If the methods of B and C are combined, using C to determine the capture mode in B, and then adopt the adaptive operation process, the crawl performance will improve.

4. Algorithm based on pattern recognition

Feature classification--Smart Hand Controller---feature extraction------effective EMG acquisition, human body

Effective EMG Acquisition: modulation (filtering, amplification) of the original EMG signal and the reliability test initiated by muscle contraction (according to the amplitude of muscle contraction EMG, statistical characteristics, etc.), and feature extraction (feature generation and dimensionality reduction) and classification algorithm (recognition and regression)

Refer to Dr. Yang Dapeng's thesis "Study on EMG control of multi-exercise mode of artificial prosthetic prosthesis"

Muscle electric Control dexterous hand (I.)