Bipedal Robots , General legged locomotion , Reinforcement Learning applications, Biomechanics, Prostheses design, Actuator design.

Bipedal Robots  

The aim of this research is to design and build bipedal robots that can achieve dynamic walking on level and rough terrain. We will adopt different designs and apply different control algorithms to the robots. Learning paradigms like reinforcement learning will also be adopted in the control algorithm. From experimentation, we wish to extract rules for general bipedal walking.

Other Research Labs

MIT Leg Laboratory (MIT, US)
Robotics/Vibration Control @ UNH (University of New Hampshire, US)
The Human Power, Biomechanics, and Robotics Laboratory (Cornell Unversity, US)
Human Engineering Laboratory (UC Berkeley, US)
Passive Walking (University of Michigan, US)
Bipedal Gait Study (Southern Methodist University, US)
Honda Humanoid Robot (Japan)
Fukuda Vachkov Laboratory (Nagoya University, Japan)
The INRIA BIP Project (France)
The Shadow Biped (UK)

General Legged Robots  

The aim of this research is to design and build legged machine that have more than two legs. We are interested to study different walking or running gaits. By building these machines, we hope to understand the mechanisms for locomotion adopted by the biological counterparts. We also wish to build machines that are able to walk dynamically and robustly over rough terrain.

Other Research Labs

Biologically Inspiried Robotics Laboratory (Case Western Reserve, Quinn, US)
Pneumatic Hexapod (University of Illinois, US)
Ambulatory Robotics Lab(McGill University, Canada)

Biomimetic Actuators  

Legged robots is required to interact with the surrounding environment. In such applications, good actuator force control is very desirable. One such actuator design is called the series elastic actuator which is adopted by MIT Leg Laboratory. It is an actuator that is connected to the external load through an elastic component. The desired force on the load is achieved by controlling the deflection of the elastic component. Series elastic actuators have many desirable properties like high bandwidth, low output impedance, shock absorption capability, etc. This research involves the design and control of the force-controlled actuator. We are interested to compare different designs, for example, rotary versus linear elastic components, steel versus elastomer elastic material, linear versus nonlinear spring. The actuator will eventually be applied to legged robots.

Other Research Labs

Leg Laboratory (MIT, US)
Biologically Inspired Actuators and sensors (NWU, US)

This research was carried out in MIT Leg Laboratory. In this research, a general control architecture for 3D dynamic walking was proposed. It is based on a divide-and-conquer approach that is assisted by learning. No dynamic models are required for the implementation.

In the approach, the walking task is divided into three subtasks: 1) to maintain body height; 2) to maintain body posture; and 3) to maintain gait stability. The first two subtasks can be achieved by simple control strategies applied to the stance leg. The third subtask is mainly effected by swing leg behavior. By partitioning the overall motion into three orthogonal planes (sagittal, frontal and transverse), the third subtask can be further divided into forward velocity control (in the sagittal plane) and lateral balance (in the frontal plane). Since there is no explicit solution for these subtasks, reinforcement learning algorithm, in particular, Q-learning with CMAC as function approximator, is used to learn the key parameters of swing leg strategy which determine the gait stability.

Based on the proposed architecture, different control algorithms are constructed and verified using the simulation models of two physical bipedal robots. The simulation analysis demonstrates that by applying an appropriate local control laws at stance ankle joints, the learning rates for the learning algorithms can be tremendously improved.

For the lateral balance, a simple control algorithm (without learning) based on the notion of symmetry is discovered. The stability of the control algorithm is achieved by applying appropriate local control law at the stance ankle joint. A dynamic model is constructed to validate this approach.

The control algorithms are also demonstrated to be general in that they are applicable to different bipeds that have different inertia and length parameters. The designer is also free to choose the walking posture and swing leg strategy. Both ``bird-like'' and ``human-like'' walking postures have been successfully implemented for the simulated bipeds.
 

1996-1997:  Rough terrain locomotion of a planar biped

This research was carried out in MIT Leg Laboratory.

One of the motivation for studying legged locomotion is that it can handle rough terrain better than wheeled vehicles. There have been successful implementation of rough terrain locomotion of legged robots which have more than two legs. However, the rough terrain implementation of biped is not common. One of the reason is that bipeds have inherent instability problem even for level ground locomotion. However, biped has the advantage in certain type of rough terrain, for eg., the stair climbing, since it requires small foot hold area for locomotion. Past research have worked on the modeling of the biped locomotion using Lagrangian mechanics, or even Newton-Euler equations. The resulting equation of motions are usually computationally demanding. Thus, many researchers tried to reduce the order of the equation of motions so that the control of the biped can be simpler and can be done in real-time. 

A control methodology called the Virtual Model Control was developed recently in the MIT Leb Lab. This control methodology requires very low computation because it does not require the computation of the inverse dynamics of the legged robots. Virtual Model Control technique has been successfully used to control 2D biped successfully for walking on level ground. The objective of my research is to apply Virtual Model Control to 2D walking biped so as to enable the biped to walk on level, uphill and downhill terrain without complex or computationally intensive algorithms. The biped also does not require extensive sensory system. 

The simulation result has demonstrated the robustness of the algorithm based on Virtual Model Control.  It has demonstrated that the blind walking of a biped over a series of unknown sloped terrain (of moderate gradient) can be achieved mainly by geometric consideration. 

We could easily adopt the same approach for stairs walking. The main difference is that the biped needs to know the transition location and the step height before planning the swing leg motion. (stairs walking simulation).   
 

1994-1995:  Distributed sequential control in automation using lonwork

This research was carried out in the Control and Mechatronics Laboratory, Department of Mechanical Engineering, National University of Singapore.

Conventional sequential control system usually consists of a central controller (for example, PLC) which has I/O modules as the interface to external input and output devices, such as solenoid operated valves, actuators, sensors, etc.  This point-to-point wiring connections usually result in messy wiring layout and it can become complex as the number of devices increases.  In recent year, most PLC systems have incorporated fieldbus (distributed system applying to low level devices) capability which consists of smart I/O devices, for example, sensors, actuators, etc. which have embedded microprocessor chip in them.  All the devices are connected by communication medium (multi-drop) which facilitate data and messages transfer between the PLC and the devices.  The purpose of this approach is to reduce the wiring cost and increase the reliability for the system. 

This research was aimed to develop a new approach for sequential control implementation using distributed system.  Linear pneumatic cylinder was used as the actuator device to demonstrate various concepts of the proposed distributed sequential control system.
The thesis has listed two target subsystems of sequential control.  One of which is the sequential control of cylinders without input from independent discrete sensors.  The other is the sequential control of cylinders with inputs from independent sensors.  Various conventional approaches for such subsystems were also  introduced.

Before proceeding to design the distributed system, an overview of distributed system architecture and certain concepts of the object-oriented development were studied.  Based on this, various approaches of implementing sequential control using distributed system were explored and a modular approach based on peer-to-peer architecture was chosen for further development.  In this peer-to-peer distributed control approach, the cylinder nodes (each controlling a cylinder system) are linked by a communication medium and capable of communicating with one another  to achieve the overall sequential control task without the need of a central controller or master-slave architecture.  Each node consists of a pre-programmed control module, which contains local I/O interface and handles the communication task.  Three other types of nodes were also designed (the sensor, alarm and user interface nodes).  The system was then implemented using the LonWorks network system to achieve the target sequential control subsystems.  An industrial automation application was simulated based on a workbench developed. 

In the peer-to-peer distributed system, the sequential control of the target systems of the research can be implemented by simply binding (logically connecting) the network variables of the nodes without the need of programming any of the nodes.  It is also easy to modify the system by changing the affected binding (logical connection).  The end-user just needs to understand the functions of the objects to implement the system.  Additional features like "Step-through" mode were also built in to assist the installation task and to help certain fault finding in the system. 

The proposed system has inherent concurrency properties and thus it can be applied to concurrent sequential control problems..  However, due to the token passing approach, the networking platform needs to have a reliable communication ability. 
Although the implemented system is applied to pneumatic cylinders throughout the research, this approach can be adopted for other discrete event control, for examples, the on-off of motor driving a conveyor or pump filling liquid into bottles. 
However, the proposed system has a few limitations and disadvantages.  Some of which are due to the limitation of the network platform used, whereas some are due to the inherent characteristics of peer-to-peer approach.
 

1990-1991:  Variable Compliance Robots:  Vibrational Studies Of A 2 DOF Flexible Limb

This research was carried out in the Control and Mechatronics Laboratory, Department of Mechanical Engineering, National University of Singapore.

Traditionally, robots arms were only capable of dealing with positioning task.  The ability for a manipulator to accomplish force as well as position control is very desirable for many tasks.  To accomplish force control, compliance is always required and tasks that require compliance have been accomplished using special compliance devices as end-effectors.  The end-point stiffness matrix of this type of robots, however, have low variability.  Researchers have studied the design of variable compliance robots.  Redundant Joint Method is one of the method that has been proposed.  In this method, the manipulator has limbs that are deliberately compliant and the end-point stiffness matrix of such robot can be varied by rotating the limbs with joints which are redundant (in kinematics sense).  Although such design will mean that robot will have variable end-point compliance, new vibration and control problems are also generated due to the fact that one or more limbs of the robot will be deliberately compliant.  The vibration may be generated during motion control or parametrically excited.  In this project, the issues of vibration were studied based on a 2 DOF setup.  The tasks were to verify the vibration problems and at the same time, tried to obtain possible solutions for the problems.

Zhou, W., Chew, C.-M., Hong, G.-S. 2007. Development of a compact double-disc magneto-rheological fluid brake. Robotica, accepted for publication.

Chew, C.-M., Hong, G.-S., Zhou, W. 2006. Series damper actuator system based on MR fluid damper. Robotica, 24(6), 668-710.

Burdet, E., Tee, K.-P., Mareeks, I., Milner, T.E., Chew, C.-M., Franklim, D.W., Osu, R., Kawato, M. 2005. Stability and motor adaptation in human arm movements. Biological Cybernetics, 94(1), 20-32.

Tee, K.-P., Burdet, E., Chew, C.-M., Milner, T.E. 2004. A model of force and impedance in human arm movements. Biological Cybernetics, 90, 368-375.

Chew, C.-M., Pratt, G.A. 2004. Frontal Plane Algorithms for Dynamics Bipedal Walking. Robotica, vol 22, 29-39.

Ho, H.-N., Chew, C.-M., Hong, G.-S., Talasila, S. 2003. Using Genetic Algorithms To Optimize Stance Ankle Behavior for Bipedal Walking. International Conference on Computational Intelligence, Robotics and Autonomous Systems, 15-18 Dec, 2003, Singapore.

Chew,C.-M.,Hong, G.-S., Wong Y.-S. 2002. Distributed Sequential Control Implementation Using Peer-to-Peer Approach. Journal of The Institution of Engineers, Singapore, Vol 42, No 5, 29-37.

Choong, E., Chew, C.-M., Poo, A.-N., Hong, G.-S. 2003. Mechanical Design of An Anthropomorphic Bipedal Robot. International Conference on Humanoid, Nanotechnology, Information Technology, Communication and Control, Environment, and Management, Mar 2003, Manila, Philippines.

Chew, C.-M., Choong, E., Poo, A.-N., Hong, G.-S. 2003. From Science Fiction to Reality - Humanoid Robots. International Conference on Humanoid, Nanotechnology, Information Technology, Communication and Control, Environment, and Management, Mar 2003, Manila, Philippines.

Chew, C.-M., Pratt, G.A. 2002. Dynamic Bipedal Walking Assisted by Learning. Robotica, vol 20, 477-491.

Pratt, J., Chew, C.-M., Torres, A., Dilworth, P., Pratt, G. 2001. Virtual model control: An intuitive approach for bipedal locomotion. International Journal of Robotics Research, 20: (2), 129-143.

Chew, C.-M., Pratt, G. A. 2001. Adaptation to Load Variations of a Planar Biped: Height Control using Robust Adaptive Control.  Robotics and Autonomous Systems , 35(1), 1-22. 

Chew, C.-M., Pratt, G.A. 2000. A General Control Architecture for Dynamic Bipedal Walking. Proceedings of IEEE International Conference on Robotics and Automation, San Francisco, California, April 2000.

Chew, C.-M., Pratt, J. E., Pratt, G. A. 1999. Blind Walking of a Planar Bipedal Robot on Sloped Terrain. Proceedings of IEEE International Conference on Robotics and Automation, Detroit, Michigan, May 1999.

Lee, Szer-Ming; Hong, Geok-Soon, Wong, Yoke-San, Chew, Chee-Meng. 1997. Directed Graph Model for Distributed Sequential Control. International Journal of Computer Applications in Technology, vol 10 no 5-6, 337-347.

Chew, C.-M.  (1998).   Height Control of a walking biped using Robust Adaptive PID Control.  Term project report for MIT course no. 2.153: Adaptive Control.

Chew, C.-M.  (1997).   Robust Adaptive Control Approach for Planar Biped Locomotion.   Term project report for MIT course no. 2.152: Nonlinear Control, Spring 1997.

Chew, C.-M.  (1997).   Virtual Model Control approach for planar biped locomotion.  Term project report for MIT course no. 2.165: Robotics and Mechatronics, Spring 1997.

Chew, C.-M. (1999).  Distributed Sequential Control In Automation Using LONWORK. M.Eng. Thesis, Department of Mechanical Engineering, National University of Singapore.

Chew, C.-M.  (1998).  Blind Walking of a Planar Biped over Sloped Terrain. MS Thesis, Department of Mechanical Engineering, Massachusetts Institute of Technology.