In this paper we present a learned alternative to the Motion Matching algorithm which retains the positive properties of Motion Matching but additionally achieves the scalability of neural-network-based generative models. Although neural-network-based generative models for character animation are capable of learning expressive, compact controllers from vast amounts of animation data, methods such as Motion Matching still remain a popular choice in the games industry due to their flexibility, predictability, low preprocessing time, and visual quality - all properties which can sometimes be difficult to achieve with neural-network-based methods. Yet, unlike neural networks, the memory usage of such methods generally scales linearly with the amount of data used, resulting in a constant trade-off between the diversity of animation which can be produced and real world production budgets. In this work we combine the benefits of both approaches and, by breaking down the Motion Matching algorithm into its individual steps, show how learned, scalable alternatives can be used to replace each operation in turn. Our final model has no need to store animation data or additional matching meta-data in memory, meaning it scales as well as existing generative models. At the same time, we preserve the behavior of Motion Matching, retaining the quality, control, and quick iteration time which are so important in the industry.

Redundant mechanical systems like humanoid robots are designed to fulfill multiple tasks at a time. A task, in velocity-resolved inverse kinematics, is a desired value for a function of the robot configuration that can be regulated with an ordinary differential equation (ODE). When facing simultaneous tasks, the corresponding equations can be grouped in a single system or, better, sorted in priority and solved each in the solutions set of higher priority tasks. This elegant framework for hierarchical task regulation has been implemented as a sequence of least-squares problems. Its limitation lies in the handling of inequality constraints, which are usually transformed into more restrictive equality constraints through potential fields. In this paper, we propose a new prioritized task-regulation framework based on a sequence of quadratic programs (QP) that removes the limitation. At the basis of the proposed algorithm, there is a study of the optimal sets resulting from the sequence of QPs. The algorithm is implemented and illustrated in simulation on the humanoid robot HRP-2.

We present a novel method for prioritizing both linear equality and inequality systems and provide one algorithm for its resolution. This algorithm can be summarized as a sequence of optimal resolutions for each linear system following their priority order. We propose an optimality criterion that is adapted to linear inequality systems and characterize the resulting optimal sets at every priority level. We have successfully applied our method to plan local motions for the humanoid robot HPR-2. We will demonstrate the validity of the method using an original scenario where linear inequality constraints are solved at lower priority than equality constraints.

A method of regrasp planning for humanoid robot manipulation is proposed. We adopt pivoting manipulation for the humanoid robot to move a bulky object without lifting in a stable and dexterous manner. In order to carry the object to a desired place, the humanoid should sometimes move through narrow areas surrounded by obstacles. We propose a roadmap multiplexing planning to allow the robot to leave the object near narrow places and to regrasp it from another position to continue carrying. We utilize visibility probabilistic roadmap (PRM) method as a preprocessing to capture the critical configurations for regrasping. Then a diffusion method is employed to plan the overall manipulation path including regrasping. The proposed method is verified through planning simulation including whole-body motions.

Abstract This paper deals with whole‐body motion planning and dynamic control for humanoid from two aspects: locomotion including manipulation and reaching. In the first part, we address a problem of simultaneous locomotion and manipulation planning that combines a geometric and kinematic motion planner with a dynamic humanoid motion generator. The second part deals with whole‐body reaching tasks by using a generalized inverse kinematics (IK) method to fully exploit the high redundancy of the humanoid robot. Through experiments using humanoid platform HRP‐2 No. 14 installed at LAAS‐CNRS, we first verify the validity of each method. An integrated experiment is then presented that unifies the both results via visual perception to execute an object‐fetching task. Copyright © 2009 John Wiley & Sons, Ltd.

This paper emphasizes on the capacity of a humanoid robot to perform tasks that are difficult for other types of robots. It deals with manipulation of bulky objects. Such tasks require complicated manipulations involving the whole-body and fine coordination between legs, arms and torso motions. We introduce here a whole-body motion planner that allows a humanoid robot to autonomously plan a pivoting strategy that accounts for the various constraints: collision avoidance, legs-arms coordination and stability control. Based on a previous result by the authors [1] proving the small-time controllability of a pivoting system, the planner is proven to inherit from the probabilistic completeness of the samplingbased motion planning method it is built on. The geometric and kinematic capacity of the proposed planner is mainly demonstrated through simulations and experiments.