human motion mapping to a robot arm with redundancy … · integration of a robotic platform able...

ARK 2014 Ljubljana, Slovenia • 1 July 2014

Human Motion Mapping to a Robot Armwith Redundancy Resolution

Fanny FICUCIELLO • Amedeo ROMANO • Vincenzo LIPPIELLOLuigi VILLANI • Bruno SICILIANO


Big challenge of humanoid robotics Provide robotic system with autonomous and dexterous skills to replicate capacity in

performing tasks which are typically executed by humans

Learn grasping and dynamic manipulation capabilities from humans Technologies

Motion capture systems Haptics Vision

Software strategies Postural synergies Neural network Dimentionality reduction techniques Redundancy resolution strategies

IntroductionHuman Motion Mapping to a Robot Arm with Redundancy Resolution 2/15


Goal Derivation of a unified framework for dynamic manipulation where the mobile nature of

the robotic system and the manipulation of non-prehensile non-rigid or deformable objects will explicitly be taken into account

Achievements Novel techniques for 3D object perception

dynamic manipulation control and reactiveplanning

Innovative mobile platform with a torso,two lightweight arms with multi-fingered hands,and a sensorized head for effective executionof complex manipulation tasks, also in thepresence of humans

The RoDyMan ProjectHuman Motion Mapping to a Robot Arm with Redundancy Resolution 3/15


Robot Dynamic ManipulationHuman Motion Mapping to a Robot Arm with Redundancy Resolution 4/15

Dynamic manipulation will be tested on an advanced demonstrator, i.e. pizza making process, where the application scenario is conceived to emulate the human ability to carry out challenging robotic tasks


Integration of a robotic platform able to acquire and transfer human-body motion to a robotic system

Learning from humans motion primitives, grasping and dynamic manipulation Xsens MVN based on inertial and magnetic measurement units used to

measure postures of human bodies 7-DoF KUKA LWR4 robot arm used for replicating human motion Cartesian impedance strategy for robot control using for position, speed and

acceleration the setpoints derived from a second-order filter processing the signals of the Xsens MVN

Redundancy used to ensure a compliant behavior of the robot elbow in order to reconfigure its position

This WorkHuman Motion Mapping to a Robot Arm with Redundancy Resolution 5/15


Xsens MVN suite does not need cameras, emitters ormarkers, simple for indoors and outdoors use (1.9Kg)

17 MTx inertial and magnetic measurement units:3D gyroscopes, accelerometers and magnetometersto obtain the position and orientation of parts of the body

Two Xbus Masters for wireless communication Sensors positioned at feet, legs, thighs, pelvis,

shoulders, sternum, head, arm, forearm and hand Real-time frequency of maximum sampling rate 120 Hz MVN Studio Software for observation of movements in 3D,

recording or exporting Tracking of dynamic motion the system estimates body segment orientation and position changes by

integration of gyroscope and accelerometer signals updated by using a biomechanical model of the human body

The initial pose between the sensor and the relative body segment is unknown: a calibration procedure, requiring measures such as the height, the arms length and the foot length of the subject, has to be performed

Experimental Set-up: Xsens MVN suiteHuman Motion Mapping to a Robot Arm with Redundancy Resolution 6/15


The robot used to replicate the human arm motion is the 7-DoF KUKA Lightweight Robot (LWR)

The kinematic redundancy, similar to the human arm, allows elbow motion while maintaining hand pose

Torque sensors, mounted in all joints, allow detecting contact and collisions for safe human–robot interaction and compliant reaction to applied external forces

The robot can be also manually guided and programmed

Experimental Set-up: 7-DoF KUKA LWRHuman Motion Mapping to a Robot Arm with Redundancy Resolution 7/15


Real time issues Measurements of joint angles of the human arm

are not provided in real time (only from recorded data)

Position and orientation of the segments (arm, forearm, hand) are provided in real time

For limited speed and smooth motion, signals from the Xsens are processed (interpolation and filtering)

Assumptions First three joints axes intersect at the center of

q2 (corresponding to the spherical joint of the human shoulder)

Joint q6 corresponds to the spherical joint of the human wrist

Human Arm vs. Robot ArmHuman Motion Mapping to a Robot Arm with Redundancy Resolution 8/15


Measurements of interest: position of the hand , forearm , and upperarm The kinematics of the human and robot are different also in terms of link dimensions To compute the wrist reference position of the robot, the versor of the human forearm and

upperarm are obtained and multiplied for the robot forearm and upperarm lenghts These quantities are summed with the vector linking joint q2 to the base of the robot

Mapping of the Human Arm MotionHuman Motion Mapping to a Robot Arm with Redundancy Resolution 9/15

Z

X

Y


Only the translational motion is considered [Khatib et al,1987]

Vector of end-effector position End-effector inertia matrix Vector of Coriolis/centrifugal forces Vector of gravitational forces Vector of friction forces Vector of control forces Vector of external forces

Operational Space Dynamic ModelHuman Motion Mapping to a Robot Arm with Redundancy Resolution 10/15


Friction and Coriolis/centrifugal forces are neglected, gravity is compensated In absence of external forces at the tip, the following impedance control guarantees tracking

of the desired end-effector pose trajectory (high control gains)

The error between the desired and the effective pose is expressed by means of the position error and orientation error expressed in terms of the quaternion

Cartesian Impedance ControlHuman Motion Mapping to a Robot Arm with Redundancy Resolution 11/15


Secondary task in the null space of the end effector is assigned as

dynamically consistent generalized inverse

is a suitable damping torque the secondary task consists in a possible reconfiguration of the arm obtained by applying forces to the robot’s body

Null-space Control for Redundancy ResolutionHuman Motion Mapping to a Robot Arm with Redundancy Resolution 12/15

no redundancy elbow up elbow down


Since the human arm positions may be out of the workspace of the robot, to generate a suitable set-point a bounding box is applied to the computed desired wrist position to impose limits on the spatial coordinates

ExperimentsHuman Motion Mapping to a Robot Arm with Redundancy Resolution 13/15


The objective of this work is to create a robotic platform to learn from human grasping and manipulation tasks Kinematic mapping algorithm to replicate the movements of human arm on an

anthropomorphic robot arm with seven degrees of freedom Cartesian impedance strategy for robot control using for position, speed and acceleration

the setpoints derived from a second-order filter processing the signals of the Xsens MVN Compliant null-space control strategy to adjust the configuration of the robot manually in

order to generate anthropomorphic configurations

Future work consists of learning primitives of motion in a low-dimensional manifold for simplified and human-like control of humanoid robots

Conclusion and Future WorkHuman Motion Mapping to a Robot Arm with Redundancy Resolution 14/15


Thanks very much for your kind attention

Fanny FICUCIELLO • Amedeo ROMANO • Vincenzo LIPPIELLOLuigi VILLANI • Bruno SICILIANO

Human Motion Mapping to a Robot Arm with Redundancy Resolution 15/15

human motion mapping to a robot arm with redundancy … · integration of a robotic platform able...

Documents