Planning and Control of Humanoid Robots
YNL has long expertise in motion planning, control and simulation. Dr. Katsu Yamane (Disney Research) studied in his Ph.D dissertation on the efficient and parallel computational algorithms for forward dynamics computation (dynamics simulation) of complex mechanical systems that change structures by contacts. His works received the best paper award of Japan Robotics Society in 2000 and the King-Sun Fu Memorial Best Transactions Paper Award, IEEE Transactions on Robotics and Automation, in 2001.
Motion planning and control have been a continuous study at YNL and contributed by many students and researches including Prof. Masafumi Okada (Tokyo Institute of Technology), Prof. Qiang Huang (Beijing Institute of Technology), Prof. Dirk Wollherr (Technical University of Munich), Dr. Fabio Zonfrilli (Procter&Gamble), Prof. Tomomichi Sugihara (Osaka University), Prof. Ko Yamamoto (University of Tokyo), Dr. Christian Ott (DLR), Dr. Oussama Kanoun (Square Enix), Dr. Carlos Felipe Santacruz-Rosero (MathWorks), and Dr. Thomas Nierhoff (Robert Bosch GmbH).
The recent motion planning study at YNL started while Prof. Quang-Cuong Pham (Nanyang Technological University) was JSPS Postdoctral Fellow. The idea of Time-Optimal Path Parameterization (TOPP) was developed and applied to Rapidly-exploring Random Trees (RRT) to grow its branches by time optimal trajectories. TOPP provided greatly simplified computation of the time-optimal trajectories.
Stephane Caron’s Ph.D dissertation (2016.03) was on Multi-Contact Whole-Body Control of Humanoid Robot. He applied RRT-TOPP for planning of critically dynamical motions of humanoid robots. A theoretical contribution of Caron’s thesis is to have proved that RRT-TOPP is probabilistically complete (namely, that it can eventually finds the solution if it ever exists). Caron then discussed the problem of motions with multi-contacts. The famous theory of Zero-Moment Point (ZMP) by the late Prof. M. Vukobratobic works for locomotion on a flat horizontal plane with only contacts on the plane. Caron generalized it by using the relationship between the wrench (force and moment at the total mass-center of the body) and the set of the feasible contact forces taking account of the frictional constraints at the contact surfaces. The relationship is represented by Gravito-Inertial Wrench Cone.
For the ZMP, the motion planning and control of biped robots has been discussed by controlling an inverted pendulum with its fulcrum in the support area on the ground. Caron showed that one can define an arbitrary plane section of Gravito-Inertial Wrench Cone as the support area on which the fulcrum of an inverted or non-inverted (if the support area is above the total center of mass) pendulum is located. Then, the motion planning and control of the multi-contact humanoid robots is discussed in general.
The second and third figures from the right are the motion planning of HRP4, a humanoid robot, with multi-contacts on non-planar surfaces using TOPP. The green areas are the plane sections of Gravito-Inertial Wrench Cone. The first from the right is the simulation of Hydra going up stairs. The first and second from the left shows the experiments of teleoperation using HRP4. The autonomy based on the firm motion planning and control theory/algorithm is the key to realize the on-board intelligence of teleoperated humanoid robot systems.