Google DeepMind releases world model Genie 4, supporting real-time interaction and native audio
Google DeepMind released the new-generation general world model Genie 4 on the evening of the 28th; it can generate and modify interactive scenes in real time based on prompts, with a single coherent world lasting about 2 minutes and visuals at about 24 frames per second.
Highlights
Rendering-free event-frame simulation accelerates reinforcement learning for autonomous robot navigation
The authors propose an analytical event simulator that directly synthesizes event streams from scene geometry and ego-motion without image rendering; generating 64x64 event observations on a single GPU takes only 17 microseconds, more than 400 times faster than advanced rendering pipelines. An event-camera reinforcement-learning obstacle-avoidance policy trained with this simulator converges in less than 30 minutes, three orders of magnitude faster than training with rendered events, and can transfer zero-shot to Isaac Sim ray-traced event simulation.
arXiv cs.ROSource ↗FTP-1: A general foundational tactile policy across tactile sensors
The research team proposes FTP-1, a general foundational tactile policy pretrained across multiple tactile sensors and embodiments, which uses heterogeneous encoders to map image, array, and state-type tactile signals into unified morphology-aware latent tokens, jointly modeled by shared tactile Transformer experts.
arXiv cs.ROSource ↗RynnWorld-Latent: A cross-embodiment world model conditioned on general latent actions
The research team proposes RynnWorld-Latent, a cross-embodiment video world model that uses RynnLAM's 3D-grounded latent actions as a shared conditioning interface and can be jointly pretrained on human interaction videos, robot demonstrations, and simulation data.
Model releases/updates
Google DeepMind releases world model Genie 4, supporting real-time interaction and native audio
Google DeepMind released the new-generation general world model Genie 4 on the evening of the 28th; it can generate and modify interactive scenes in real time based on prompts, with a single coherent world lasting about 2 minutes and visuals at about 24 frames per second.
Industry news
Yaskawa Electric and GAZOO Racing to introduce AI autonomous welding robot technology into GR KART manufacturing
Yaskawa Electric and Toyota GAZOO Racing jointly developed AI-autonomous robot production technology and began introducing it at the Gamagori GR KART workshop in September 2026. The technology is based on the SFA method (Sequence Freezing Arc Welding) jointly developed by both parties in 2024, uses MOTOMAN NEXT robots equipped with GPUs and AI, and can autonomously recognize the state of pipe frames, complete welding, quality confirmation, and correction, and comprehensively analyze appearance images, welding sounds, and various data.
Paper research
Rendering-free event-frame simulation accelerates reinforcement learning for autonomous robot navigation
The authors propose an analytical event simulator that directly synthesizes event streams from scene geometry and ego-motion without image rendering; generating 64x64 event observations on a single GPU takes only 17 microseconds, more than 400 times faster than advanced rendering pipelines. An event-camera reinforcement-learning obstacle-avoidance policy trained with this simulator converges in less than 30 minutes, three orders of magnitude faster than training with rendered events, and can transfer zero-shot to Isaac Sim ray-traced event simulation.
arXiv cs.ROSource ↗FTP-1: A general foundational tactile policy across tactile sensors
The research team proposes FTP-1, a general foundational tactile policy pretrained across multiple tactile sensors and embodiments, which uses heterogeneous encoders to map image, array, and state-type tactile signals into unified morphology-aware latent tokens, jointly modeled by shared tactile Transformer experts.
arXiv cs.ROSource ↗RynnWorld-Latent: A cross-embodiment world model conditioned on general latent actions
The research team proposes RynnWorld-Latent, a cross-embodiment video world model that uses RynnLAM's 3D-grounded latent actions as a shared conditioning interface and can be jointly pretrained on human interaction videos, robot demonstrations, and simulation data.
WorldOracle releases action-conditioned world model benchmark; predictions of seven models deviate from real execution
Princeton University, the University of Texas at Austin, the University of Toronto, NVIDIA, and AgiBot jointly launched WorldOracle, an action-conditioned world model benchmark that compares model predictions with real execution using the same initial frame and action sequence.
Brief updates
Two open-source datasets from GenRobot.AI enter the top five of the Hugging Face robotics dataset trending list
GenRobot.AI announced that its two open-source datasets, Gen-HumanWholeBody and Gen-HumanEgo, both entered the top five of the Hugging Face robotics dataset trending list. Among them, Gen-HumanWholeBody is a first-person whole-body manipulation dataset containing 600 hours of whole-body manipulation data, 16,976 real scenes, and 1,281 human skills, with whole-body MPJPE of about 3 cm; the authors state that it targets whole-body motion learning, human-to-robot transfer, and embodied foundation model research.