Athlete Robot

2005–2010

Introduction

The human skeleton consists of many bones supported by ligaments and cartilage. Muscles, tendons, and joints between bones provide mechanical structure for movement. The musculoskeletal system gives animals the ability to move actively in a huge variety of environments. The musculoskeletal system has unique mechanisms such as bi-articular muscle, antagonistic muscle pairs and muscle-tendon elasticity. It appears that the human body is very different from conventional robot mechanisms. To understand the role of the musculoskeletal body in dynamic locomotion, we developed a Athlete Robot with an artificial musculoskeletal system.
running
Fig.1 Athlete Robot running
The results shows that the robot can perform bipedal balancing, vertical jumping of 0.5m and soft landing from a drop height of 1.0m. We also demonstrate the Athlete Robot running for five steps after a catapult launch.

Robot Design

Active, Dynamic, and Tough

The Athlete Robot is a bipedal robot based on a biomechanical approach. There are two types of robot with different mechanisms on the lower limb. The Athlete Robot weighs about 10 kg and the body height is about 1.2 m. The pneumatic muscle actuators used for the system allows agile and compliant movement of the robot.
prototype and final version
Fig.2 Two versions of robots: a prototype robot with ankle joint (left) and a final version of the Athlete Robot with blade foot(right)
The robot is equipped with a miniature computer, valves, and accumulator. Electric power and compressed air are provided from external equipment. A contact sensor for each foot, a pressure sensor for each muscle, and an inertia measurement unit (IMU) are available for measurement.

Artificial Musculoskeletal System

The artificial musculoskeletal system of the robot is based on the anatomical structure of the human. The McKibben type pneumatic artificial muscle is used for the actuation. The robot has mono-articular muscle groups and bi-articular muscle groups corresponding to the major muscles of the lower extremities of a human body. Each muscle group consists of several muscles.
artificial muscle and biological muscle
Fig.3 Ppneumatic artificial muscles mimic a biological muscle
We determined the parameters for the artificial musculoskeletal system based on the data in biomechanics. The tendons in the human lower leg play the crucial roles of energy storage and impact absorption. Biomechanical studies indicate that the human lower leg behaves in the same manner as spring during dynamic movement. Therefore, we designed the robot foot as an elastic blade.
muscle layout
Fig.4 Layout of the muscles. The abbreviations are, Gmin: gluteus minimus muscle; ADD: adductor muscles; Gmax: gluteus maximus muscle; IL: iliopsoas muscle; HAM: hamstrings; RF: rectus femoris muscle; VAS: vastus muscles.

Other Techniques

The skeletal frame consists mostly of polymer parts. The polymer bearings, nylon joint parts, and carbon FRP tubes contribute to lightweight and high-impact durable skeletal frame. We employ selective laser sintering (SLS) technology to manufacture the polymer parts with complex form.
plastic parts
Fig.5 plastic parts

Control

Computer Simulation

We use a computer simulation in order to investigate appropriate parameters for the robot. The physical parameters of the body parts are calculated from the 3DCAD models. The curving shape of the blade foot used for detecting ground contact in the simulator are also imported from 3DCAD. We modeled the elastic bending of the blade foot as a translational spring.
running snapshots
Fig.7 Computer simulation of the robot

Muscle Activation Pattern

The motor command for running is derived from simulation-based optimization. The combination of constrained random sampling and hill-climbing optimization improves a muscle activation pattern step by step. We use reconstructed human EMG data to get initial patterns for the optimization. The constraints from human skills facilitate effective motor learning.
muscle activation pattern
Fig.6 Generating motor command based on muscle activation pattern

Results

Jumping

The vertical jumping is a simple movement which widely used to study power, skills, and characteristics of the musculo-tendon. Our experiments confirmed that the robot can reach jump heights of 0.5 m. Simple step signals with different timings are used for the motor command.
vertical jump
Fig.8 Vertical jumping

Passive Soft Landing

The robot can land softly from one meter drop by exploiting the anti-gravity muscles and its compliance. The innner pressure of each muscle is regulated at the constant pressure. Landing task is particularly difficult for the robot which is driven by the geared motors because of the large instantaneous forces and short duration.
bounce from a drop
Fig.9 Soft landing from a drop height of 1.0m

Passive Bouncing

The musculoskeletal leg can use preset stiffness to control posture predictively. Here, we achieve passive control of the bouncing using preset stiffness. The stiffness of the leg is expressed as ellipsoid and its gradient of long axis. The results shows that we can control the direction of the bouncing both fall forward and fall backward.
bounce from a drop
Fig.10 Soft landing from a drop height of 1.0m

Bipedal Balancing

We performed the balance control during bipedal stance by the PID control of each joint. Although the movement of the center of mass is relatively greater than for a stiff robot, the observed sway is similar to human movement reported in biomechanics research. We also tested two types of balancing strategy: knee strategy and ankle strategy.
plastic parts
Fig.11 Bipedal balancing and the cyclic oscilation of the center of mass.

Sprinting

We demonstrated that the robot can run five steps at a velocity of 2.42m/s (8.7km/h) using an open-loop motor command. The ground reaction force during running presented a bell-shaped profile without impulse despite using simplified muscle activation pattern. The results showed the significant contribution made by an elastic lower limb and predictive motor command in agile legged locomotion.
running snapshots
Fig.12 Athlete Robot running 5 steps

Ball Kicking

The direct drive mechanism with series muscle-tendon allows high-speed joint movements. We demonstrate the kicking motion with official soccer ball. The maximum angular velocity of the knee joint is 10.5rad/s, which is comparable speed to human motion during running.
ball kicking
Fig.13 Kicking a soccer ball

References

Contribution

Ryuma Niiyama organized the project, under the supervision of Professor Yasuo Kuniyoshi. Ryuma Niiyama designed all electric circuits, mechanisms, and controllers. The simulation study of the robot is the work of Satoshi Nishikawa and Ryuma Niiyama. Special thanks to Kazuya Shida who helped me with the real robot experiments.

Papers

Ryuma Niiyama, Satoshi Nishikawa and Yasuo Kuniyoshi
A Biomechanical Approach to Open-loop Bipedal Running with a Musculoskeletal Athlete Robot
Advanced Robotics, Vol.26, No.3–4, 2012. (in press)
BibTeX

@ARTICLE{Niiyama2010_OpenloopBipedalRunning-with-AthleteRobot,
author = {Ryuma Niiyama and Satoshi Nishikawa and Yasuo Kuniyoshi},
title = {A Biomechanical Approach to Open-loop Bipedal Running with a Musculoskeletal Athlete Robot},
journal = {Advanced Robotics},
volume = {26},
number = {3},
year = {2012},
}

Ryuma Niiyama and Yasuo Kuniyoshi
Design Principle Based on Maximum Output Force Profile for a Musculoskeletal Robot
Industrial Robot: An International Journal, Vol.37, No.3, pp.250–255, 2010.
PDF
BibTeX

@ARTICLE{Niiyama2010_MOF-Profile-for-MusculoskeletalRobot,
author = {Ryuma Niiyama and Yasuo Kuniyoshi},
title = {Design Principle Based on Maximum Output Force Profile for a Musculoskeletal Robot},
journal = {Industrial Robot: An International Journal},
volume = {37},
number = {3},
pages = {250--255},
year = {2010},
}

Ryuma Niiyama, Satoshi Nishikawa and Yasuo Kuniyoshi
Athlete Robot with Applied Human Muscle Activation Patterns for Bipedal Running
In Proceedings of the IEEE-RAS International Conference on Humanoid Robots (Humanoids 2010), pp.498–503, Nashville TN, USA, Dec. 2010.
PDF
BibTeX

@INPROCEEDINGS{Niiyama2010_Humanoids2010_AthleteRobot,
author = {Ryuma Niiyama and Satoshi Nishikawa and Yasuo Kuniyoshi},
title = {Athlete Robot with Applied Human Muscle Activation Patterns for Bipedal Running},
booktitle = {Proc. IEEE-RAS Int. Conf. on Humanoid Robots ({Humanoids 2010})},
year = {2010},
pages = {498--503},
address = {Nashville, Tennessee USA},
month = {Dec.},
}

Ryuma Niiyama and Yasuo Kuniyoshi
Design of a Musculoskeletal Athlete Robot: A Biomechanical Approach
In Proceedings of the 12th International Conference on Climbing and Walking Robots and the Support Technologies for Mobile Machines (CLAWAR 2009), pp.173–180, Istanbul, Turkey, Sept. 2009.
PDF
BibTeX

@INCOLLECTION{Niiyama2009_MusculoskeletalAthleteRobot-BiomechanicalApproach_book,
author = {Ryuma Niiyama and Yasuo Kuniyoshi},
title = {Design of a Musculoskeletal Athlete Robot: A Biomechanical Approach},
booktitle = {Mobile Robitcs: Solutions and Challenges, Proc. of the 12th Int. Conf. on Climbing and Walking Robots ({CLAWAR 2009})},
publisher = {World Scientific Publishing},
year = {2009},
pages = {173--180},
isbn = {978-981-4291-26-2},
}

Ryuma Niiyama and Yasuo Kuniyoshi
A Pneumatic Biped with an Artificial Musculoskeletal System
In Proceedings of the 4th International Symposium on Adaptive Motion of Animals and Machines (AMAM 2008), pp.80–81, Cleveland, Ohio USA, June 2008.
PDF
BibTeX

@INPROCEEDINGS{Niiyama2008_PneumaticBiped-with-ArtificialMusculoskeletalSystem,
author = {Ryuma Niiyama and Yasuo Kuniyoshi},
title = {A Pneumatic Biped with an Artificial Musculoskeletal System},
booktitle = {Proc. 4th Int. Symposium on Adaptive Motion of Animals and Machines ({AMAM 2008})},
year = {2008},
pages = {80--81},
address = {Cleveland, Ohio USA},
month = {June},
}