Trajectory Planning and Control of Serial Linked Robotic Arm for Fruit Picking Robot using Reinforcement Learning

Project Title

Trajectory Planning and Control of Serial Linked Robotic Arm for Fruit Picking Robot using Reinforcement Learning

Project Area of Specialization RoboticsProject Summary

Agricultural yield is very important and essential need for human beings because they directly depend on it for essential living. Especially, fruits are typically rich in nourishment thus bought by every household and it requires a continuous supply and production to satisfy the demand of the growing world population. Fruit picking is a labor-intensive, time-consuming, and costly task in fruit production, therefore, it is indispensable to develop an automatic fruit harvesting robot such as Pick and place robotswhich are commonly used in modern manufacturing environments. Primary benefits of pick and place are speed and consistency with reliable performance. These robots can handle complex repetitive tasks while freeing up human workers. Robots can be customized to meet specific production requirements. Fruit picking is one of the most time-consuming, cost demanding, and laborious tasks in fruit production. At present, the main methods employed in fruit picking robots for fruit detection are based on machine learning algorithms combined with machine vision. A lot of work has been done on collision avoidance such as Artificial Potential Field (APF) method. These methods can be used to plan collision-free paths, but these are inefficient. Collisions can damage both the robot and the picked fruit thus minimizing its performance. In this project, a fast and robust collision-free path planning method based on Deep Reinforcement Learning (DRL) for continuous state spaces in dynamic environment will be used.

Project Objectives

The objectives of this project are to:

• Design a reinforcement learning based controller for a robotic arm functioning in a continuous workspace.
• Make the robot efficient enough to avoid collisions with good grasping ability.
• Develop a prototype of fruit picking robot.

Project Implementation Method

Fruit picking is a process in which a robot uses an end-effector to grasp a fruit to minimize human effort, work load, increase speed, accuracy and efficiency. To achieve this goal, we will design a controller such that our serial linked robotic arm can effectively perform collision free tasks in a dynamic environment. In this project we will use Raspberry Pi, Servo motors, Arduino, Shaft encoders, RGB cameras and a manipulator. First, we will perform image acquisition by using RGB camera which will retrieve images from the environment. Then these will be used for image processing, which will involve reduction of noise and enhancement of input images. Next, we will perform segmentation, to differentiate between fruit and branches as it is appropriate and provides more detail about the target fruit and background. Based on this, some picking points will generate for collision- free path planning so that robot can approach to the targeted fruit along these waypoints without colliding with branches.

To accomplish this, we will use a fast and robust collision-free path planning method based on Model-free reinforcement learning used for continuous state spaces in dynamic environment. Once training is complete, our robot will know about the optimal estimation of pose that is position and orientation of end- effector, it will go for the most efficient and shortest path by stretching its arm. After reaching near the target, it will grasp the fruit using pressure sensor with its end-effector and pluck it away.

Benefits of the Project

1. Consistency.
2. Accuracy and speed.
3. Decreases labor cost.
4. Reduces food wastage. 
5. Increases Financial sustainability.

Technical Details of Final Deliverable

In this project a serial linked fruit picking robotic arm will be developed, equipped with RL based controller, which can pick and place fruits efficiently in a continuous environment. The design the controller using RL learning will enable the manipulator to perform all the tasks of detection and grasping the target fruit without any encounter to collisions.

Final Deliverable of the Project HW/SW integrated systemCore Industry AgricultureOther Industries Food Core Technology RoboticsOther Technologies Artificial Intelligence(AI)Sustainable Development Goals Decent Work and Economic Growth, Industry, Innovation and InfrastructureRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	80000
Servo motors	Equipment	8	4500	36000
Arduino Uno	Equipment	4	850	3400
Raspbarry Pi SBC	Equipment	1	8500	8500
RGB camera	Equipment	1	10500	10500
Connection wires	Equipment	100	20	2000
Aluminum serial link manipulator	Equipment	1	9600	9600
Structure Development Labour (Welding and cutting etc)	Miscellaneous	1	6000	6000
Printing	Miscellaneous	1	4000	4000