Smart Piezo-Resistive Sensing And Experience Based Control For Soft Robot Grasping

Soft robots have the tendency to achieve shape adaptable and complex motions due to their multi degree of freedoms(DOF's).These DOF's are provided enough space for soft robots to perform the tasks in the open loop control and achieved the notable success as compare to rigid robots.These artificial soft robots are dominantly inspired from humans but relatively do not exhibit the two most important pillars of bio-inspiration i.e.smart embedded sensing and decision making,so from this thesis work we move the soft robots a step towards decision-making and intelligent sensing.The haptic perception and position feedback enables the robot to understand the real-world interactions in a better way.However,the most important thing is that the perception demands a subject and object to deal in a qualitative manner.In this thesis,we have worked on 2-D soft robots i.e.grippers as the subject and the different objects of variable stiffness,different weights and geometries as objects through which data is perceived.In comparison with the human hands that are capable of providing the haptic and position feedback,which is not present in soft robotic grippers and this lag limits the soft grippers to work in the open loop control domain.In this work,we provide a soft gripper that is applicable for the close loop control applications.We design the cable driven soft gripper in a way that it is capable of providing the haptic,position feedback,and fulfills the criteria of haptic and position perception and its utilization in the close loop control tasks.Our sensors differ from previous works in terms of materials,design and fabrication perspective.Previously,the pressure sensors used for soft grippers are not compatible with the compliance of the soft actuators because they were attached externally on the soft body and limit the motion of the actuators,so we devised a way to create pressure sensors inside the soft body which are purely soft in nature and fulfils the concept of soft haptic perception.On the other end the position sensors,designed in this work are purely made up of soft fabrics,which outclass the previous fabric sensors in terms of their bulky structure.They are also effective for the soft robot continuous mobility applications because some previous position sensors design is only useful with the discrete motions of the soft robot.We also perform the sensors characterization and grasping experiments to prove that our sensors give repeatable results during the grasping tasks.The other part of this work is focused on the development of optimal control strategy that is purely closed and based on the concept of decision-making.The grasps with soft actuators are stochastic rather than deterministic because of the soft actuators extra compliance and due to the object geometry,deformation factor and weight parameters.The stochastic grasps in unstructured environment creates a probabilistic effect.For this problem,we proposed a novel Q-learning experience strategy that divide the grasping experiences into multiple events and then optimized the grasp by minimizing the cost function.The soft finger behavioral model and the grasp model are specially modeled for maintaining the probabilistic events and are the main ingredient of the grasp strategy.The soft gripper equipped with the position feedback module provides the data used in the modeling.The output of this strategy is that the soft robotic agent learns to grasp the weighted and deformable objects with ease and in control loop manner.In addition to this,the data set is made efficient by objects of different weights and stiffness.In the end,a qualitative analysis is presented to prove the unique optimization methodology.