HRP-2 reproducing a human slalom - the whole process

Abstract

The work presented in this paper deals with the generation of trajectories for humanoid robots imitating human gaits captured with a motion capture system. Once the human motion is recorded, this one is modified to be adapted to the robot morphology. The proposed method could be used for human-like robots of different sizes and masses. The generated gaits are closed to the humans ones while respecting the robot balance and the floor contacts. First the human joint angles are computed and directly applied to the robot kinematic model as a marionette. From this virtual motion, the trajectories of both feet and of the Zero Moment Point (ZMP) are generated respecting the constraints of floor contact and balance control. Moreover a strategy to avoid automatically feet collisions during single supports is detailed. From this data, an optimization algorithm is used to compute the joint angles of the robot while minimizing their variations. In this paper we present the results of a slalom transposed for the robot HRP-2 (Kawada Indutries, Inc).

A computational model for synchronous motion imitation by humans: The mirror controller applied on stepping motions

“…Thus, these methods have to solve several problems such as singularities, selfcollisions, joint dynamic limits and of course stability in the case of whole-body imitation. For example, in all the previous works targeting the same objective of imitating walking or dancing motions [2], [8], [9], [11], [12], the contact phases between the feet and the ground need to be extracted first from the sensor signals. It is only then that the steps' lengths and timings can be deduced, so the length of a step can be identified only once it has been completed.…”

“…7 shows that the imitator successfully find the current step position of the demonstrator (mean error 2.6mm) without any delay even with a higher noise level (10cm/s 2 of standard deviation). 2…”

An auto-adaptable algorithm to generate human-like locomotion for different humanoid robots based on motion capture data

“…That is why the study is limited to a quasistatic balance approach, which one, as the pendulum method, consists in controlling the CoM location. For example, a dynamic walking reproducing a human slalom was also generated for HRP-2 [19].…”