MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202641124618 A) filed by Srm University on October 17, 2026, for System And Method For Generating Physically Plausible Human Motion Data Using A Graph-Action- Masked Transformer.
Inventors include Sahoo, Abhisek; Reddy, Chinthalacheruvu Avinash; Reddy, Karri Deepak; Sai Anish, Gunda Mokshith; Mishra, Sugyan Kumar; and Mishra, Sambit Kumar.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: ABSTRACT The present disclosure relates to a system (100) and method (200) for generating physically plausible human motion data. Body keypoints are detected (204) in performance videos and segmented (206) into motion clips of N frames. A directed motion graph (128) is constructed (208) with edges between sequential clips or between clips whose boundary poses satisfy a pose-distance threshold and an angular-velocity threshold. A graph attention encoder, trained with a contrastive loss, computes (210) latent node embeddings. An autoregressive transformer generates (212) a node sequence, a graph action mask (130) derived from the outgoing edges of the previously selected node assigning negative infinity to logits of unconnected nodes, so that only valid transitions are sampled. The clips are assembled (214), refined (216) by projecting keypoints onto calibrated bone-segment lengths, and output (218) for rendering an animated avatar on a display device (126).
Disclaimer: Curated by HT Syndication.
