My Research

 

Humanoid robot research

My research interests in humanoid robotics include trajectory generation, balancing and bilateral control (Haptic) of humanoid robots. In particular I have developed a model predictive controller that can in realtime, plan center-of-mass (COM) trajectories in 3 dimensions (emphasis on variable COM height, often planners assume constant height) as well as foot positions (reactive stepping).

I am also interested in bilateral force/position control of humanoid robot’s arms as I believe teleoperation of a humanoid robot is a practical and useful technology, one problem though is that arm forces can make the legs loose balance thus my work employs reactive stepping, trajectory compensation (robot will lean forward to allow greater arm forces) and state-dependent arm force constraints (which are haptically reflected back to the master)

 

 

One limitation of the above mentioned force constrained bilateral teleoperation method is that it does not allow large impulsive forces to be applied. Yet the robot will not neccesarily fall just because the arms apply a large force for a short timespan. By creating an appropriate COM position dependent arm velocity constraint, the arms can generate an impulsive force while gauranteeing that the master operator will not make the robot fall. Here is a video demonstration of the proposed concept.

I also think that current ZMP feedback based balancing techniques have an unnecessarily weak force rejection capacity. This happens because such methods allow the COM to move forward even with weak disturbances but the closer the COM comes to the edge of the support polygon, the weaker the rejection force becomes. My proposed method applies the maximum permissible rejection force without letting the feet tip over (ZMP stays within support polygon), this prevents the COM from moving towards unnecessarily and thus prevents an unnecessarily weak rejection force. My method also limits the COM velocity to prevent overshoot.

 

 

DSC_0680

By the way, the small robot used in my experiments was developed by myself (I call it “urn00b”) based on the Robotis MX-64 actuators which are capable of being controlled at a high feedback rate (1Khz if multiple communication channels are used but I use 200Hz for simplicity) and torque control.

Finally, I am also working on a motion planner that can plan joint space trajectories (not simplified workspace, i.e not COM trajectories) and constrain the ZMP position while also respecting joint position and velocity limits.

 

 

Particle Swarm optimisation on Graphics Processors

swingup

I was also interested in motion planning by using a particle swarm optimisation (PSO). Since PSO is a highly paralell problem, it is well suited for computation on a GPU.

 

 

 

 

 

Mechanisms

omniCopter

Another research interest of mine is mechanical mechanisms. One idea was to equip a UAV with three arms that are equipped with Omni-wheels to either grip an object or to allow the UAV to drive on the ground (which may be more energy efficient in some situations).

3dof3dofAnother idea was to make a 3 degree of freedom joint that has the peculiar property of having all its motors fixed to the base link. The idea is to reduce wiring failure by having the actuators fixed in place. On the right are a couple ideas towards this goal, the one is a parallel link mechanism with a universal joint for the yaw motion and the other is a differential mechanism.

 

 

 

 

Electrostatic actuators

I am also currently developing a three-phase electrostatic motor (not electromagnetic). Usually these types of motors are considered weak but I believe they can exceed the torque-to-weight ratio of their electromagnetic counterparts, further research will clarify this matter.

emotor2emotormultiplier

 

 Posted by at 7:16 am