. 2022 Jan 3;8:749274. doi: 10.3389/frobt.2021.749274

TABLE 14.

Animation and virtual environments.

Reference	AI Algorithm/Characteristics	Data Acquisition/Inputs	Task
Feigl et al. (2020)	THR, COR, SVM and BiLSTM, tested	N = 6, head-mounted accelerometer data	Motion reconstruction
Feigl et al. (2020)	- COR has the best accuracy for real-time VR applications (low delay)	N = 6, head-mounted accelerometer data	Gait phase detection
Bergamin et al. (2019)	DReCon: motion matching and deep RL	Unstructured motion data from mocap	Real-time physics-based character control for video games
Bergamin et al. (2019)	- responsive to user demands, natural-looking. Trained on flat terrain	Unstructured motion data from mocap	Real-time physics-based character control for video games
Peng et al. (2018b)	OpenPose/HMR and DRL	Simulated character model and YouTube video clip	Learning dynamic physics-based character controllers from video clips
Peng et al. (2018b)	- Learning from inexpensive video clips, robust	Simulated character model and YouTube video clip
Peng et al. (2018a)	DeepMimic: DRL	Character model, kinematic reference motion from video clip	Physics-based character controllers from video clips
Peng et al. (2018a)	- Diverse skills/terrains/morphologies, realistic response to perturbations	Character model, kinematic reference motion from video clip	Physics-based character controllers from video clips
Huang et al. (2018)	SMPL body model and BiLSTM	6 IMUs	3D human pose reconstruction from a sparse set of IMUs
Huang et al. (2018)	- Useful when camera-based data is not available due to occlusion, fast motion, etc	6 IMUs	3D human pose reconstruction from a sparse set of IMUs
Holden et al. (2016)	CAE	CMU Motion Capture Database ³	Unsupervised learning of a human motion manifold
Holden et al. (2016)	- Capable of fixing corrupt data, filling in missing data, motion interpolation along the manifold, and motion comparison	CMU Motion Capture Database ³	Unsupervised learning of a human motion manifold
Huang et al. (2015)	SMG and part-based Laplacian deformation	Three 4DPC datasets ⁴	A data-driven approach for animating 4DPC character models
Huang et al. (2015)	- Simultaneously captures both motion and appearance for video-like quality	Three 4DPC datasets ⁴	A data-driven approach for animating 4DPC character models
Ding and Fan, (2015)	Multilayer JGPMs/topologically constrained GPLVMs	CMU Motion Capture Database + Simulated data	Human gait modeling
Ding and Fan, (2015)	- diversity of walking styles, motion interpolation, reconstruction, and filtering	CMU Motion Capture Database + Simulated data	Human gait modeling
Alvarez-Alvarez et al. (2012)	FFSM with automatic learning of the fuzzy KB by GA	N = 20	Human gait modeling
Alvarez-Alvarez et al. (2012)	- Fuzzy states and transitions are still defined by experts, interpretable, generalizes well for each person’s gait	Accelerometer attached to the belt	Human gait modeling

Legend: Threshold Based Method (THR), Pearson Correlation-based Method (COR), Data-Driven Responsive Control (DReCon), Human Mesh Recovery (HMR), Deep Deterministic Policy Gradient (DDPG), Skinned Multi-Person Linear (SMPL) as in (Loper et al., 2015), 4D Performance Capture (4DPC), Surface Motion Graphs (SMGs), Carnegie Mellon University (CMU), Joint Gait-Pose Manifolds (JGPMs), Fuzzy Finite State Machines (FFSM), Knowledge Base (KB).

Datasets: ³ http://mocap.cs.cmu.edu/ ⁴ http://cvssp.org/cvssp3d.