阅读说明：本技术 一种基于智能演进运动技能学习的仿生四足机器人 (Bionic quadruped robot based on intelligent evolution motor skill learning ) 是由 朱晓庆 张晔鹏 陈璐 孙泽文 鲍昕 周夏阳 杨晓蓬 于 2021-08-30 设计创作，主要内容包括：本发明公开了一种基于智能演进运动技能学习的仿生四足机器人,包括作为四足机器人主体框架和四肢,四个腿部单元沿机身的两侧对称布置,相对其主体框架中垂面呈镜像对称。红外线传感器和摄像头与控制板连接,电池与控制板连接,控制板与驱动装置连接。四足机器人头部单元包含：红外线传感器和摄像头,并与主体内部的控制板连接。该四足机器人拥有八自由度,通过协作控制,轻便灵活。仿生四足机器人步态学习采用智能演进运动技能学习,通过深度神经网络、奖励引导完善运动技能学习。(The invention discloses a bionic quadruped robot based on intelligent evolution motor skill learning, which comprises a main body frame of the quadruped robot and four limbs, wherein four leg units are symmetrically arranged along two sides of a robot body and are in mirror symmetry relative to the middle vertical plane of the main body frame. The infrared sensor and the camera are connected with the control panel, the battery is connected with the control panel, and the control panel is connected with the driving device. The four-legged robot head unit includes: infrared ray sensor and camera to be connected with the inside control panel of main part. The quadruped robot has eight degrees of freedom, and is light and flexible through cooperative control. The gait learning of the bionic quadruped robot adopts intelligent evolution motor skill learning, and the motor skill learning is perfected through deep neural network and reward guidance.)

1. The utility model provides a bionical four-footed robot based on intelligence evolution motor skill study which characterized in that: comprises a machine body (1), legs (2), a knee bracket (3), a head unit (4), a camera (5), an infrared sensor (6), a battery (7), a control panel (8) and a driving device (9); the four legs (2) are symmetrically arranged along two sides of the machine body (1); the battery (7), the control panel (8) and the driving device (9) are arranged in the machine body (1);

the camera (5) and the infrared sensor (6) are connected with the control panel (8), the battery (7) is connected with the control panel (8), the control panel (8) is connected with the driving device (9), and the driving device (9) is connected with the legs (2); when the robot works, the four legs (2) are driven to rotate by one driving device (9) respectively.

2. The bionic quadruped robot based on intelligent evolution motor skill learning of claim 1, characterized in that: the top of upper limbs (11) is connected through connecting piece (18) with fuselage (1), upper limbs (11), fuselage (1) and connecting piece (18) are connected through first pivot (13), upper limbs (11) are connected through second pivot (15) with low limbs (12), foot (17) set up in the bottom of low limbs (12), first brushless motor (14) control upper limbs (11) around first pivot (13) swing, second brushless motor (16) control low limbs (12) around second pivot (15) swing.

3. The bionic quadruped robot based on intelligent evolution motor skill learning of claim 1, wherein the legs are characterized in that: the ratio of upper and lower limbs is 1:1.2, the same leg ratio as for the horse is adopted; an upper limb hollow part (19) is adopted in an upper limb (11), the upper and lower parts of the upper limb hollow part (19) are semicircular, the left side and the right side of the upper limb hollow part are linear tetrahedrons, the hollow proportion accounts for 50%, a lower limb hollow part (20) is adopted in a lower limb (12), a foot hollow part (21) is adopted in a foot (17), the lower limb hollow part (20) and the foot hollow part (21) respectively occupy most areas of the lower limb (12) and the foot (17) in consideration of the weights of the lower limb and the foot, and the hollow proportion accounts for 90%; the bottom of the foot (17) adopts a gear-shaped structure, so that the adhesion to the ground is increased; the second rotating shaft (15) is surrounded by the knee support (3), and can limit the angle between the lower limb (12) and the upper limb (11) to be 160 degrees at most and 70 degrees at least, thereby playing a certain buffer effect.

4. The bionic quadruped robot based on intelligent evolution motor skill learning of claim 1, characterized in that: the device consists of a camera (5), an infrared sensor (6) and an external structure (10); the obtained picture information and video information and the data information obtained by the infrared sensor (6) are transmitted to the control panel (8) through the camera (5) for processing, and then the result is fed back to the driving device (9) to drive the four legs (2) to carry out different tasks.

5. The mechanical gait synchronous hexapod robot according to claim 1, characterized in that: the four legs are divided into a left forelimb 23, a right forelimb 24, a left hind limb 25 and a right hind limb 26 which are respectively driven by the driving device 9;

the walking process of the robot is as follows:

1) when the robot runs straight, the left front limb 23 and the right rear limb 26 form a group, the right front limb 24 and the left rear limb 25 form a group, the two groups alternately swing forwards to advance in a running state, and the movement conditions are completely symmetrical;

2) when backing, the left forelimb 23 and the right hind limb 26 are in a group, the right forelimb 24 and the left hind limb 25 are in a group, the two groups alternately swing backwards, backing with walking gait, and the movement conditions are completely symmetrical;

3) when the robot turns left in place, the left front limb 23 and the right rear limb 26 form a group, the left front limb 23 swings forwards greatly, the right rear limb 26 swings forwards in a small amplitude, the right front limb 24 and the left rear limb 25 swing forwards in a small amplitude, and the two groups swing alternately, so that the robot turns left in place;

4) in the pivot right turning, the left front limb 23 and the right rear limb 26 are in a group and swing forwards with small amplitude, the right front limb 24 and the left rear limb 25 are in a group, the right front limb 24 swings forwards with large amplitude, the left rear limb 25 swings forwards with small amplitude, and the two groups of alternative swing robots realize pivot right turning.

6. The bionic quadruped robot based on intelligent evolution motor skill learning of claim 1, characterized in that: the gait learning of the bionic quadruped robot adopts a method of intelligent evolution motor skill learning to learn motor skills, and the robot can realize autonomous learning and self-improvement through the intelligent evolution motor skill learning; the control steps of the main program flow of the system are as follows,

step 1, initializing a quadruped robot to enable a gait state to return to zero;

step 2, setting a reward guide function type, judging whether the type is speed or stability, if so, carrying out the next step, and if not, switching to another type for judgment;

step 3, according to a preset threshold value, through a deep neural network, the quadruped robot performs gait learning, performs cyclic model training, and strengthens the learning effect until the requirements are met;

step 4, outputting the learning result as the gait of the robot, and ending the program;

different learning directions, high speed, low speed and stability are provided by selecting different types of reward guide functions in any mode, and different training models are called to realize corresponding functions.

Technical Field

The invention discloses a bionic quadruped robot based on intelligent evolution motor skill learning, and belongs to the technical field of bionic robot design.

Background

In recent years, the robot technology is rapidly developed, the bionic quadruped robot is used as one of the classifications, quadruped animals are used as templates, the bionic quadruped robot can well cope with complex environments or special environments, and the bionic quadruped robot has higher and higher application value under dangerous and complex conditions such as scientific research, medical treatment, geological survey, disaster rescue and the like. At present, the main difficulties of the bionic quadruped robot are that the manufacturing cost is high and the control level does not reach the required precision, so that the design of the quadruped robot is light and flexible, can identify the surrounding environment, and has a firm structure and high stability, which is the key point of the current research.

Disclosure of Invention

The invention aims to provide a bionic quadruped robot based on intelligent evolution motor skill learning, which has the functions of walking on flat ground, turning, environment perception and the like. .

In order to achieve the above object, the present invention provides a bionic quadruped robot, the appearance of which is shown in fig. 1, and the robot comprises a body 1, legs 2, a knee support 3, a head unit 4, a camera 5 and an infrared sensor 6; the four legs 2 are symmetrically arranged along two sides of the machine body 1; the internal structure of the robot is shown in fig. 2, and a battery 7, a control board 8 and a driving device 9 are installed in the body 1.

As shown in fig. 2, the camera 5 and the infrared sensor 6 in the head unit 4 are connected to a control board 8, the battery 7 is connected to the control board 8, the control board 8 is connected to a driving device 9, and the driving device 9 is connected to the leg 2. When the robot works, the four legs 2 are driven to rotate by one driving device 9.

The leg 2 of the robot is configured as shown in fig. 3, and includes a knee support 3, an upper limb 11, a lower limb 12, a first rotating shaft 13, a first brushless motor 14, a second rotating shaft 15, a second brushless motor 16, a foot 17, and a link 18. The top of upper limbs 11 passes through connecting piece 18 with fuselage 1 to be connected, and upper limbs 11, fuselage 1 and connecting piece 18 are connected through first pivot 13, and upper limbs 11 passes through second pivot 15 with low limbs 12 to be connected, and sufficient 17 sets up the bottom at low limbs 12, and first brushless motor 14 control upper limbs 11 is around the swing of first pivot 13, and second brushless motor 16 control low limbs 12 is around the swing of second pivot 15.

As shown in fig. 4, the robot leg 2 is seen in a side view, in which an upper limb hollow 19 is used for the upper limb 11, an upper limb hollow 20 is used for the lower limb 12, and a foot hollow 21 is used for the foot 17. The second rotation shaft 15 is surrounded by the knee brace 3, and can restrict the angle between the lower limb 12 and the upper limb 11 to a maximum of 180 ° and a minimum of 45 °, thereby achieving a certain cushioning effect.

The head unit 4 of the robot is composed of a camera 5, an infrared sensor 6, and an external structure 10, as shown in fig. 5. The obtained picture information and video information and the data information obtained by the infrared sensor 6 are transported to the control board 8 through the camera 5 for processing, and then the results are fed back to the driving device 9 to drive the four legs 2 to perform different tasks.

The gait learning of the bionic quadruped robot adopts a method of intelligent evolution motor skill learning to learn motor skills, the motor skill indexes comprise speed and stability in various aspects, the robot learning direction is determined by setting a reward guide function for classification, and then the robot is trained to achieve the indexes through a deep neural network. The control steps of the main program flow of the system are as follows,

step 1, initializing the quadruped robot and enabling the gait state to return to zero.

And 2, setting the type of the reward guide function, judging whether the type is speed or stability, if so, carrying out the next step, and if not, switching to another type for judgment.

And 3, according to a preset threshold value, performing gait learning by the quadruped robot through a deep neural network, performing cycle model training, and strengthening the learning effect until the requirements are met.

And 4, outputting the learning result as the gait of the robot, and ending the program.

Different learning directions, high speed, low speed and stability can be provided by selecting different types of the reward guide function in any mode, and different training models are called to realize corresponding functions. Compared with the prior art, the invention has the following innovations:

1. the structure that adopts camera and sensor to use is carried out the three-dimensional reconstruction of environment to transmit the processing result and drive leg motion for drive arrangement, reduced the error signal that the error brought

2. The leg structure design adopts partial hollow design, can effectively reduce the dead weight of the leg, keeps the stability and the firmness of the leg joint, and reduces the damage rate of the leg structure

3. The robot has strong stability when walking. The tail end of the leg is made of rubber materials with grooves on the surfaces, so that the contact area and time of the leg of the robot and the ground are increased, and the probability of sinking and slipping can be reduced.

4. By adopting intelligent evolution motor skill learning and deep neural network training, the robot can realize autonomous learning and self-improvement and can improve a plurality of motor skill indexes such as speed, stability and the like.

Drawings

Fig. 1 is an overall appearance of the robot of fig. 1.

Fig. 2 internal structure of the robot.

Figure 3 is a schematic view of the leg structure.

Fig. 4 is a side view of a leg.

Figure 5 front view of the robot.

Figure 6 top view of the robot.

Fig. 7 intelligent evolving motor skill learning flow diagram.

Detailed Description

As shown in fig. 6, the robot has four legs, namely, a left front leg 23, a right front leg 24, a left rear leg 25, and a right rear leg 26, which are driven by the driving device 9.

The walking process of the robot is as follows:

1) when the robot runs straight, the left front limb 23 and the right rear limb 26 form a group, the right front limb 24 and the left rear limb 25 form a group, and the two groups alternately swing forwards to advance in a running state, so that the movement conditions are completely symmetrical.

2) In the case of backpedaling, the left forelimb 23 and right hindlimb 26 are in one group and the right forelimb 24 and left hindlimb 25 are in one group, the two groups alternately swing backwards to backpedal with a walking gait and the motion is completely symmetrical.

3) When the robot turns left in place, the left front limb 23 and the right rear limb 26 form a group, the left front limb 23 swings forwards greatly, the right rear limb 26 swings forwards with small amplitude, the right front limb 24 and the left rear limb 25 swing forwards with small amplitude simultaneously, and the two groups swing alternately, so that the robot can turn left in place.

4) When the robot turns right in place, the left front limb 23 and the right rear limb 26 form a group and swing forwards by a small amplitude at the same time, the right front limb 24 and the left rear limb 25 form a group, the right front limb 24 swings forwards by a large amplitude, the left rear limb 25 swings forwards by a small amplitude, and the two groups of alternative swing robots can realize turning right in place.