Page 35 - ASME DSCC 2015 Program

Technical Program

on the System-Theoretic Passivity Properties of a Hill Muscle Model effect of parallel compliance on Stability of robotic hands With Series
Invited session paper. DSCC2015-9817 Elastic Actuators
Hanz Richter, Antonie J. van den Bogert, Cleveland State University, Invited session paper. DSCC2015-9917
Cleveland, OH, United States Prashant Rao, Taylor niehues, Ashish Deshpande, The University of Texas
at Austin, Austin, TX, United States
The paper describes passivity-related input-output properties of a human
muscle and tendon system given by a Hill type dynamic model. For a model Compliance is a key requirement for safe interactions with the environment
having muscle contraction velocity as the input and force as the output, it for any robot. It has been well established that the human body exploits var-
is shown that the system is passive during the concentric phase. Also, it is ious arrangements of compliance such as series compliance (musculo-ten-
shown that a negative strict passivity margin exists for the eccentric phase if don units) and parallel compliance (joint capsules and ligament complex) to
the length and lengthening velocity of the contractile element are assumed achieve robust and graceful interaction with the environment. Mechanical
bounded. Estimates of this margin are given by means of two alternative compliance can be similarly arranged in robotic joints in series or parallel to
formulas. Further, it is shown that the mapping from contraction velocity to actuators. The effects of such arrangements on the closed loop properties
deviation from equilibrium force is passive in both concentric and eccentric of robotic joints such as stability, disturbance rejection and tracking perfor-
phases. The paper discusses how these findings and the passivity theorem mance have been analyzed separately, but their combined effects have not
can be used to design controllers for a machine coupled to the muscle mod- been studied. We present a detailed analysis on the combined effects of
el by feedback interconnection. The simple case of a proportionalderivative series and parallel arrangements of compliance on low inertia robotic joints.
force feedback regulator is considered as an example. A simulation exam- Our analysis shows the stability limitations of achievable joint stiffness due
ple is given where the transient response of the coupled system crosses the to series compliance and the subsequent increase in the stable upper limit
eccentric region. of achievable joint stiffness by addition of parallel compliance. We provide
guidelines towards designing compliance to improve the stability and perfor-
learning Contracting nonlinear Dynamics from Human Demonstration
mance of low-inertia robotic joints, which can be applied to the improvement
for Robot Motion Planning
of robotic hands performing grasping and manipulation tasks. We validate
Invited session paper. DSCC2015-9870
our analysis by means of an experimental platform and discuss the various
Harish Ravichandar, Ashwin Dani, University of Connecticut, Storrs, CT, characteristics and the effects of both arrangements of compliance on
United States
robotic hands.
In this paper, we present an algorithm to learn the dynamics of human arm
Supernumerary Robotic fingers as a Therapeutic Device for
motion from the data collected from human actions. Learning the motion
Hemiparetic Patients
plans from human demonstrations is essential in making robot program-
Invited session paper. DSCC2015-9945
ming possible by non-expert programmers as well as realizing human-robot
collaboration. The highly complex human reaching motion is generated by Teddy ort, faye Wu, nicholas Hensel, Haruhiko Asada, Massachusetts
Institute of Technology, Cambridge, MA, United States
a stable closed-loop dynamical system. To capture the complexity a neural
network (NN) is used to represent the dynamics of the human motion states. Patients with hemiparesis often have limited functionality in the left or right
The trajectories of arm generated by humans for reaching to a place are hand. The standard therapeutic approach requires the patient to attempt
contracting towards the goal location from various initial conditions with built to make use of the weak hand even though it is not functionally capable,
in obstacle avoidance. To take into consideration the contracting nature of which can result in feelings of frustration. Furthermore, hemiparetic patients
the human motion dynamics the unknown motion model is learned using a also face challenges in completing many bimanual tasks, for example walker
NN subject to contraction analysis constraints. To learn the NN parameters manipulation, that are critical to patients’ independence and quality of life.
an optimization problem is formulated by relaxing the non-convex contrac- A prototype therapeutic device with two supernumerary robotic fingers was
tion constraints to Linear matrix inequality (LMI) constraints. Sequential Qua- used to determine if robotic fingers could functionally assist a human in the
dratic Programming (SQP) is used to solve the optimization problem subject performance of bimanual tasks by observing the pose of the healthy hand.
to the LMI constraints. For obstacle avoidance a negative gradient of the Specific focus was placed on the identification of a straightforward control
repulsive potential function is added to the learned contracting NN model. routine which would allow a patient to carry out simple manipulation tasks
Experiments are conducted on Baxter robot platform to show that the robot with some intermittent input from a therapist. Part of this routine involved
can generate reaching paths from the contracting NN dynamics learned allowing a patient to switch between active and inactive monitoring of hand
from human demonstrated data recorded using Microsoft Kinect sensor. The position, resulting in additional manipulation capabilities. The prototype suc-
algorithm is able to adapt to situations for which the demonstrations are not cessfully enabled a test subject to complete various bimanual tasks using
available, e.g., an obstacle placed in the path. the robotic fingers in place of normal hand motions. From these results, it
is clear that the device could allow a hemiparetic patient to complete tasks
which would previously have been impossible to perform.

35