• Antagonistic Actuation
• Impedance Control
• Internal Force
• Manipulability
• Non-linear transmission
• Physical Human Robot Interaction
• Rollarticular Joint
• Simulataneous Control of stiffness and motion
• Stiffness Controllability
• Tendon Driven Arm
• Variable Stiffness/Impedance Actuation

This dissertation presents an approach toward development of a serially connected robotic arm meant for physical Human Robot Interaction. The motivation starts from the effort in incorporating a property of the manipulator, attributing to “Intrinsically Safe” robot. In the advent of the thesis work, a review of state-of-the-art is carried out in finding a suitable metric (or, a collection of safety measures) to quantify impact/collision related physical safety, concerning physical interaction between robot and human. The dominant property of a robot contributing to potential safety hazards are its inertia and speed of operation. With conventional robots for safety critical physically interactive applications, this speed is kept below rather a low value leading to loss of productivity. Away from the conventional approach, the thesis-work then proceeds to follow a design concept - design for safety first and control it for performance. The concept follows an inertia attenuating approach in an optimal way in doing a trade off
between safety (impact) and performance.

In this respect of seeing safety (due to impact) of human as a function of effective inertia, nonetheless, restricts the size of inertia of robot links to be low compared to human. The identified fact of maximum energy transfer at impact requiring equal inertias of colliding bodies put this restriction to make inertia attenuating approaches effective. This aspect had not yet been completely explained in literature; attenuating inertia does not help in increasing relative safety, unless the robot link inertia is low compared to the mass of an object, whose safety is in concern and then the total inertia (from motor) attains an equal value.

Reducing the effect of the sources of impact hazards, a safe design culminates in assembling a lightweight structure, possessing compliant components, for example, by putting considerable flexibilities in transmissions (apart from soft cover/skin). The loss of performance in placing compliance is recovered by varying the transmissio impedance/stiffness. The Variable Impedance Actuation (VIA) evolves. The VIA, or, a class of it, the Variable Stiffness Actuation(VSA) is an outcome of a time optimal task planning problem for strict safety constraint with variable joint impedance during motion. Performance of a flexible jointed robot is enhanced with variable impedance/stiffness and obtained are two desired trajectories for a
specified task, one for velocity and the second for stiffness, leading to Stiff-&-Slow and Soft-&-Fast paradigms.

The thesis-work follows the method of antagonistic actuation through nonlinear transmissions in putting forth stiffness variation and motion control. In this activity, a new non-linear transmission element is developed from a first principle learnt from biological muscle characteristics, which leads to a spring with exponential force-length characteristic. A systematic general method is developed to realize mechanically any arbitrary spring of continuous monotonic force-length function, using profile cam.

The dissertation proceeds to address design and analysis of tendon structures for serially connected arm - the notion of Tendon Manipulablity and Stiffness Controllability are introduced and elaborated. Role of joint-space redundancy is considered for task-space-stiffness control. Finally, design of a base-actuated tendon driven system with 2-degrees of joint-space-redundancy is presented; the serial manipulator arm possess a 3-dof intersecting shoulder and two elbows. Concept of tendon driven differential shoulder is delineated in the context. The first partial prototype of the fore arm consisting of two Rollarticular elbows is developed in laboratory and assembled.

For a single-degree-of-freedom joint, there are two actuators in antagonism, while in minimally manipulable n-degrees-of-freedom arm there can be (n + 1) to 2n actuators with varied degree of stiffness controllability in joint and taskspace. All these attributes tend to converge to emulation of a biological system - the human arm. The requirement of being lightweight, stiffnes variable and biomorphic almost uniquely converge to a tendon driven arm. A new rolling type hingeless joint, developed in the thesis work adds to the biomorphic design of the robot joint. This, named as Rollarticular joint, introduces a completely new joint type, unlike the conventional robotic joints found in literature and industry. The new joint is then formalized with robotic convention and terminology.

Rollarticular joint offers a joint particularly suitable for tendon actuation. Use of Rollarticular mechanism makes it possible to decouple the tendon movements while passing through successive joints from distal end to base (assuming all the joints possess the same mechanism). This pertains to stiffness controllability - even the minimally manipulable tendon driven arm becomes completely joint stiffness controllable, unlike coupled n + 1 or 2n arrangements with hinge joints.

The final part of the dissertation considers the control aspects. An elaborate review of control of flexible joint robots (fixed compliance) is presented. A very recent method of feedback-linearization based control for variable stiffness actuation is studied in comparison with a new method proposed here. The thesis proposes
a force sensor (tendon tension sensors) based impedance controller for the simultaneous control of motion and stiffness for task following(planned velocity and stiffness trajectories for stiff-slow and soft-fast motion). Unlike most of the other approaches found in literature, the proposed method is sensor based (position, velocity states and force), which avoids parameter uncertainties of model based
approaches and becomes less computationally involved. The approach assumes an ideal behaviour of the sensors (at present ignores force sensor signal noise).

The notion of dependency of input-settable stiffness on the null-space component
of antagonistic tendon forces is identified; in this method, stiffness is estimated in
terms of the sensed internal force components. This reduces further online computational
burden. This novel approach, based on the internal tendon forces, hence implements a controller for simultaneous motion and stiffness trajectory following in a very simple manner. Identifying input-dependent stiffness lying in the null space of antagonistic system and applying them in stiffness control is one of the new contributions of this thesis.