Model-Based Design And Development Of A Table Tennis Training Robot

Aims of the Study

Describe about the Model-Based Design and Development.

In the field of robotics as well as artificial intelligence, Robot table tennis is a challenging domain. Based on the concept of robotics, it requires reliable perception as well as control regarding artificial intelligence [8]. It needs fast decision making in order to determine the best motion to hit the ball. In the year 1988, the first robot ping-pong player is presented who is capable of participating against humans as well as machines.

This paper focuses on modeling human motor skills such as Table tennis, which has the fundamental works of complex tasks and it needs precise control. It provides the aims and objectives of the model-based design and development of table tennis training robot. This paper reviews four articles based on the research study. Research methodology is given to identify the methods used to propose the technical approach and prepare a plan of the research.

The aim of this research study is to model and develop the design of a table tennis robot to play in the environment. It is required to become familiar with the different approaches, which enables the robot to adapt to the task and improve the performance.

To identify the four stages of table tennis ball rally

To innovate a low price table tennis robot with a distinctive control system

To develop a new design and concept to solve the design problems of table tennis robot

To identify the use of Reinforcement Learning which enables the robot to improve the performance

To become familiar with the Mixture of Motor Primitive framework

Playing in table tennis is a vital and demanding task for the robots as the robots have deficiencies to perceive the environment to the hardware limitations that limit the actions [3]. In most of the research problems, robot table tennis is used as the standard task for high-speed vision as well as fast movement generation. The robots require a repertoire of actions to cover the potential hitting points in the entire workspace [7]. The biometric robot player employs three actions such as duration, amplitude as well as ultimate goal position. As the robots are faster as compared to human beings, then the robot player suffers from hardware limitations such as torque as well as acceleration limits [9]. Four stages are categorized during playing of expert players such as:

Awaiting stage: The ball progresses towards the rival who strikes it reverse towards the opponent. At this stage, the racket goes downwards [8]. At the consummation stage, the racket will be on the flat parallel to the table plane.

Objectives

Preparation stage: In this stage, the ball comes towards the player; it has as of now conceded the opponent as well as strike the table. The racket moves in reverse to compose to hit.

Hitting stage: The ball moves towards the point where the buyer catches it [8]. The racket goes towards the ball until it hits in a roundabout expansion.

Finishing stage: At the last stage, the ball is on the arrival way towards the adversary though the racket moves upwards to the halting position.

Figure 1: Four Stages of Table tennis ball rally

(Source: [8])

From the four stages, it is seen that the robot gains sufficient time to initiate as well as execute a hitting movement, where the robots anticipate about the intention of the human opponent [3]. The prediction of the intention of the human opponent is realized by modeling the intention directs the dynamics of hitting movements. In the scenario of table tennis training robot, the Intention-Driven Dynamics Model (IDDM) leads to an online algorithm that gives a time series of observations and predicts the target and intention of the human player [7]. However, the robot needs some certain period in order to execute its hitting movement. In any sports, there are two players to play the game but if one player feels to play but other may not [10]. In order to overcome the uncertainty, it has designed a table tennis robot. It has a single degree of freedom. The robots used to throw the ball automatically within the time interval of 1 second where the players play against the robot. Existing table tennis robot is designed using a microcontroller as well as a programming unit [6]. In order to reduce the cost as well as complexity, a ping-pong robot is designed using mechanical parts.

Toward the start of each stage, an engine primitive requires being picked taking into account the environmental stimulus. These primitives establish the order as well as the moment of the muscles contraction [4]. Sensory information is required to modify the motor primitives in order to generate rapid corrections due to change in environmental demands recognized in the table tennis. [2] used an innovative state automation to represent the computational realization of the model. In order to understand the four phases of table tennis ball rally, the table tennis system detects the presence of the ball as well as senses its position p_b. By utilizing the advancement technique approach, the issues of learning robot abilities are ordered as the robots begins with unique preparing information set accomplished from the exhibits from which it ponders an underlying arrangement [3]. The first goal of this method is to accomplish by imitation learning as well as the second goal is to require reinforcement learning. Therefore, learning of model has required for improvement the execution.

Literature Review

Proposed Technical Approach

Model based design is used for development of dynamic systems such as control systems. In this design, system model is used to develop through three stages such as design, implementation as well as testing. With Simulink, the user moves towards idealized linear models to realistic non-linear models. Simulink provides a graphical editor to build models such as block diagrams [9]. As the models are hierarchical, therefore it is build using both top-down as well as bottom-up approaches. In order to define the model, mathematical integration models are used by interactively in Simulink [5]. The use of Reinforcement Learning (RL) enables the robot in order to adapt to the task, improves the performance while it requires very less domain knowledge. The combination of the Reinforcement Learning with the imitation learning enables the human users in order to provide domain knowledge to the RL algorithm [1]. The experts on the task provide the domain knowledge if they have less expertise in the robotics [8].

Imitation learning relies on the inverse reinforcement learning while the RL users different dynamic framework in the robot table tennis that is capable of returning the ball to a success rate of 94 percent. Once the hitting point of the ball is estimated, the decision module determines the hitting motion of the ball from parameterized hitting motions studied from human expressions [12]. The motion primitives are combined with a single motion, which depends on the circumstances such as the position of the ball as well as velocity and the pose of the robot. The weights are adjusted using the reinforcement learning [11]. A mixture of Motor Primitive framework is a novel approach that is evaluated in the approach of setting up of table tennis where the researcher can improve the performance of the player.

In this research proposal, the secondary data is used. The data are collected from the four articles. In this research project, the research methodology plays an important role to build the framework for this research. The philosophy of the research refers to the nature of its reality that is investigated. The first approach that is the deductive approach helps to do the evaluation of different types of existing approaches with support to an analysis of data.

Work Breakdown Structure

Work Breakdown	Activities in the work breakdown
	Model-based design and development of table tennis training robot
1.0	Selection of the research topic
1.1	Identify the research aims and objectives
2.0	Collecting data to research on the topic
3.0	Creating a layout of the research report
4.0	Preparation of review the four articles
4.1	Review the four stages of tennis ball rally
4.2	Study the Intention-Driven Dynamics Model
4.3	Review the designing process of Existing table tennis robot
5.0	Formation of research methods
5.1	Proposing of technical approach
5.2	Analyzing the technical progress data
6.0	Findings of the data
7.0	Preparation of final work

Figure 2: Gantt chart for project planning

(Source: Created by author)

Conclusion

It is concluded that the imitation, as well as reinforcement learning methods, use to take in wide assortments of engine aptitudes. It ranges from the easy errands such as ball paddling, dart games up to playing the robot table tennis. In this research report, the researcher evaluated some proposed framework with human-robot table tennis games, in which the bottleneck has a limited amount of time for the robot to execute the hitting movement. Movement initiation requires an early conclusion on the type of action such as forehand as well as backhand movement before the opponent strikes the ball. The design of a smart ping-pong robot overcomes the problem faced by other roots, which are designed in the electronic platform. This robot gives high performance with low cost as compared to other robots. Since the design is in the mechanical platform. Therefore the problem occurs in electronic platform should be avoided.

References

[1]P. Du, “Application of Mechanical Analysis in Table Tennis Flight Simulation Based On the Differential Model”, Information Technology J., vol. 12, no. 14, pp. 2990-2994, 2013.

[2]J. Zou, Q. Liu and Z. Yang, “Development of a Moodle course for schoolchildren’s table tennis learning based on Competence Motivation Theory: Its effectiveness in comparison to traditional training method”, Computers & Education, vol. 59, no. 2, pp. 294-303, 2012.

[3]R. Ma and H. Xing, “Table Tennis Training Intelligent Assessment System Based on Placement Identification”, AMM, vol. 440, pp. 341-345, 2013.

[4]X. Zhang, “Design and Manufacture of a Forehand Attack Exercising Device for Teaching and Training of Table Tennis”, AMM, vol. 509, pp. 96-100, 2014.

[5]Y. Chen, “Development of a Two-Dimensional Laser Array Sensor to Detect the Ball Location for Table Tennis Training”, adv sci lett, vol. 8, no. 1, pp. 257-261, 2012.

[6]Z. Wang, A. Boularias, K. Mülling, B. Schölkopf and J. Peters, “Anticipatory action selection for human–robot table tennis”, Artificial Intelligence, 2014.

[7]K. Mulling, J. Kober, O. Kroemer and J. Peters, “Learning to select and generalize striking movements in robot table tennis”, The International Journal of Robotics Research, vol. 32, no. 3, pp. 263-279, 2013.

[8]H. Koppula and A. Saxena, “Anticipating Human Activities Using Object Affordances for Reactive Robotic Response”, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 38, no. 1, pp. 14-29, 2016.

[9]J. Li, X. Chen, Q. Huang, X. Chen, Z. Yu and Y. Duo, “Designation and Control of Landing Points for Competitive Robotic Table Tennis”, International Journal of Advanced Robotic Systems, p. 1, 2015.

[10]R. Silva, F. Melo and M. Veloso, “Towards table tennis with a quadrotor autonomous learning robot and onboard vision”, In Intelligent Robots and Systems (IROS), pp. 649-655, 2015.

[11]L. WANG, J. CAO and C. HAN, “Superquadrics Model-based 3D Object Localization Algorithm”,Robot, vol. 35, no. 4, p. 439, 2013.

[12]L. Hantrais, “Methodological pluralism in international comparative research”, International Journal of Social Research Methodology, vol. 17, no. 2, pp. 133-145, 2014.