Design of Robotic Hand and Tactile Sensor for Robotic Grasping and Manipulation

面向抓取與操縱的機械手和傳感器設計

Student thesis: Doctoral Thesis

View graph of relations

Author(s)

Related Research Unit(s)

Detail(s)

Awarding Institution
Supervisors/Advisors
  • Yajing SHEN (Supervisor)
  • Jia Pan (External person) (External Co-Supervisor)
Award date29 Aug 2022

Abstract

Robotic grasping and dexterous manipulation at a human-like level are essential for a fully functional robot. However, excellent hardware design, e.g., robotic hand and tactile sensors, fully supporting the work mentioned above, is still an open and complex problem that challenges the designer's knowledge and experience. To date, no robotic hand poses a human-like grasping and manipulation capability, and no robust tactile sensor embedded in the robot’s finger performs like the human finger skin. Thus, this thesis focuses on designing robotic hands and tactile sensors, aiming to assist various robotic grasping and manipulation applications.

This thesis first designs anthropomorphic fingers employing a biomimetic actuation mechanism, which accomplishes independent movement of interphalangeal (IP) joints and metacarpal (MCP) joints. The proposed design simultaneously reduces the number of actuators without losing the DOFs necessary for achieving high-quality grasping postures in the cable-driven anthropomorphic hand design. We further exploit this actuator mechanism in the suggested robotic finger design and accomplish all the GRASP taxonomy's 33 static and stable grasping postures.

Second, we present a computational framework for automatic optimal robotic hand design based on reinforcement learning (RL), which considers the desired grasping tasks, grasp control strategies, and performance quality measures. The RL-based framework intends to develop finger joints with different types and link lengths at different positions from null. Then, the reward function for such a growing action relies on various quality indexes from the generated robotic hand, aiming to perform the desired grasping tasks under the expected control strategies. To demonstrate our framework’s effectiveness, we set the desired task to grasp objects of three primitive shapes, i.e., box, cylinder, and sphere, with predefined hand positions and the ways to close fingers to grasp each object. The grasping force closure condition, quantitative stability indexes, energy consumption, and some penalty terms assemble the reward function. We demonstrate that the proposed framework can automatically generate capable robotic hands through simulation and practical prototype experiments. Potential factors affecting the framework’s output deserving further exploration are also discussed.

In addition to robotic hands, tactile sensors are also essential for grasping and manipulation. Due to the merits of chemical inertness and immunity to electromagnetic interference, lightweight, small size, and softness, we design an optical waveguide tactile sensor using two layers of cores, where one has a uniform width and the other an incremental width. It is deduced and verified that the contact force magnitude can be derived from the light power loss in the uniform-width core, while the contact position can be derived from the light power loss in the other core and the estimated force. This dual-core design allows a single waveguide to measure the contact force magnitude and position simultaneously. A hardware experiment demonstrates the design’s effectiveness on a two-finger gripper in an assembly task, revealing that the dual-core waveguide achieves 2 mm spatial resolution and 0.1 N sensitivity.

In summary, this thesis presents one anthropomorphic hand design, one computational framework for automatic optimal robotic hand design, and a new waveguide design for tactile sensors in robotic fingers. These designs provide partially solid preconditions for robotic grasping and manipulation, which are open problems worth investigating.