GigaBrain-0.5M is a compact Vision-Language-Action model that leverages world model-based reinforcement learning to achieve efficient and generalizable robot manipulation.

GigaBrain-0 is a VLA foundation model trained with large-scale data generated by world models (e.g., video generation, real2real/human/view transfer, sim2real).

GigaWorld-0 is a unified world-model framework built as a data engine for VLA learning. It combines GigaWorld-0-Video for controllable, temporally coherent video generation and GigaWorld-0-3D for geometry- and physics-consistent synthesis.

SwiftVLA enhances lightweight Vision–Language–Action models with efficient 4D spatiotemporal understanding, offering exceptional performance on Orin edge devices with significantly reduced latency and memory usage.

ReconDreamer-RL integrates video diffusion priors into scene reconstruction to improve reinforcement learning for end-to-end autonomous driving.

GigaWorld-Policy is an action-centered World–Action Model (WAM) that learns 2D pixel–action dynamics with optional video generation. It runs 9× faster than the leading WAM baseline Motus, improves task success rates by 7%, and outperforms π_0.5 by 95% on RoboTwin 2.0.

WonderTurbo generates interactive 3D worlds in record time using a novel framework that enables fast and realistic world generation.

ReconDreamer enhances autonomous driving scene reconstruction by integrating world model knowledge and employing a progressive data update strategy, effectively addressing the challenges of rendering complex maneuvers.

FA-YOLO is a new framework for target detection that improves performance in adverse weather by enhancing image features and boundary clarity.

A human-machine gaming system developed at low cost using Raspberry Pi, robotic arm, and camera, involving lightweight target detection models and gaming algorithms.

WonderFree proposes a method to enhance novel view synthesis quality and cross-view consistency, providing an improved experience in 3D scene exploration with enhanced realism and detail.

WonderTurbo++ is the first real-time interactive 3D world generation framework capable of generating novel perspectives of 3D scenes within 0.58 seconds, achieving up to 20× speedup over baseline methods while maintaining spatial consistency and high-quality outputs.

GigaWorld-Policy is an action-centered World–Action Model (WAM) that learns 2D pixel–action dynamics with optional video generation. It runs 9× faster than the leading WAM baseline Motus, improves task success rates by 7%, and outperforms π_0.5 by 95% on RoboTwin 2.0.