Transition Motion Synthesis for Object Interaction based on Learning Transition Strategies

Abstract

In this study, we focus on developing a motion synthesis framework that generates a natural transition motion between two different behaviours to interact with a moving object. Specifically, the proposed framework generates the transition motion, bridging from a locomotive behaviour to an object interaction behaviour. And, the transition motion should adapt to the spatiotemporal variation of the target object in an online manner, so as to naturally connect the behaviours. To solve this issue, we propose a framework that combines a regression model and a transition motion planner. The neural network-based regression model estimates the reference transition strategy to guide the reference pattern of the transitioning, adapted to the varying situation. The transition motion planner reconstructs the transition motion based on the reference pattern while considering dynamic constraints that avoid the footskate and interaction constraints. The proposed framework is validated to synthesize various transition motions while adapting to the spatio-temporal variation of the object by using object grasping motion, and athletic motions in soccer.

Simplified Physical Model‐based Balance‐preserving Motion Re‐targeting for Physical Simulation

“…We compared the proposed framework with the baseline framework that re‐used the footstep pattern of the original motion analysis [KLVDP20, HPKI22]. Concisely, the baseline framework did not include the re‐targeting parameter learning module adapting to the target character model.…”

“…They [ALL*20] extended a skeleton‐aware network for deep motion re‐targeting in which it defines primal skeleton structure to share the same feature between the source and the target. Different from these previous works, we have extended a previous parameter learning model [HPKI22] that synthesizes transition motions by learning footstep planning parameters. We have applied the deep neural network structure to learn the parameter mapping between the source model and the target model.…”

“…The basic structure of the neural network model follows the previous work [HPKI22]. The number of hidden layers is two with 512 neurons each.…”

Efficient modeling of liquid splashing via graph neural networks with adaptive filter and aggregator fusion

“…Inspired by Zhou et al ’s (2022) work integrating the Gustafson–Kessel clustering algorithm with an adaptive recurrent fuzzy neural network, FEGNS achieves remarkable performance, attaining 92% accuracy on simulations involving over 10,000 liquid particles, while realizing a 60% reduction in computational time compared to conventional techniques (Skoulikaris and Piliouras, 2023; Hewage et al , 2024). This achievement marks a advancement toward efficient fluid simulation, pioneering the integration of advanced adaptive filtering and aggregator fusion techniques into the field of liquid splashing modeling (Hwang and Ishii, 2024; Burgard et al , 2023).…”