Five years ago we asked: can an AI agent outrace the world's best #GranTurismo drivers? The answer became a @nature.com cover, a game feature and a research frontier that's still open. #GTSophy, five years on ↓ bit.ly/4vHyxpF
@sonyai.bsky.social
Since our @nature.com paper, #Ace won against seven pro table tennis players, including two-time Olympic silver medalist Miu Hirano, and world No. 26 Miyuu Kihara. Learn how improvements across #RL, hardware, #perception, and more helped close the gap: bit.ly/440oyPM
Sony AI is presenting 10 papers at #CVPR2026 in Denver, covering generative modeling, face aging, video-to-audio synthesis, 3D scene understanding, and evolving-world perception. 👉 bit.ly/4obvwuD
#SPARC: a single racing policy that holds up across 100+ unseen vehicles — no vehicle-specific tuning needed. 👉 Learn more: arxiv.org/pdf/2511.09737
@alicex84.bsky.social, Global Head of AI Governance at Sony and Lead Research Scientist @sonyai.bsky.social, writes in @time.com: data nihilism is driving one of the greatest wealth transfers in history. The solution is ethical innovation built on consent and shared value.
Are We Entering the Age of Data Nihilism?
People feel powerless to control their data, which could widen inequality in the AI era warns Alice Xiang.
bit.ly
#ICASSP2026 Sony AI is presenting 11 papers in Barcelona this May, covering #music understanding, #generativeaudio, audio-visual alignment, and #dataquality. Each addresses a different gap between what #audio AI systems produce and what they actually understand. 👉 Learn more: bit.ly/4dsZPsX
Lorenzo Servadei, Head of AI for Chip Design at Sony AI, speaks with Silicon Semiconductor on AI-powered EDA, #GENIE-ASI for analog subcircuit identification, and #Schemato for human-readable schematics. Read the interview: siliconsemiconductor.net/magazine/686 #SonyAI #ChipDesign
A defining month for Sony AI: Project #Ace reached the cover of @nature.com, a new framework on #AI companionship, a full slate at #ICLR 2026, and a feature on the future of #chip design. 🔗 Read the April Month in Review: bit.ly/3OEuutX
AI has largely existed in virtual spaces. Ace was built to change that—and today, that research was published in @nature.com. Michael Spranger, President of Sony AI, explains what it actually required.
Why can't you just program a robot to play table tennis? Peter Dürr and Peter Stone explain, and the answer gets to the heart of what makes physical AI hard.
Over several years, we’ve contributed #research to @iclr-conf.bsky.social exploring how #machinelearning models are trained, interpreted, and applied. This year’s papers span #multimodallearning, #diffusion, interpretability, and theory, with open #code and demos. 🔗 bit.ly/4eFumVs
What happens when a ball hits the net? In table tennis, net contact creates unpredictable trajectories. For Ace, Sony AI's physical AI research system, these rare events were one of the hardest real-world conditions to address.
How does Ace—Sony AI's robotic AI research system—plan a serve? Published in @nature.com, Ace is the first robot to beat a professional athlete in a physical sport. Before a serve, Ace predicts how spin will affect the ball's flight and bounce to guide real-time execution.
How does Ace read spin in real time? Published in @nature.com, Ace is the first robot to beat a professional athlete in a physical sport. Its Gaze Control System tracks spin during fast rallies, helping it interpret the game as it unfolds.
The first robot ping-pong prototype appeared in 1983. For more than forty years, no machine could see, decide, and act within the window required to compete with elite players. That changes today.
Every autonomous robot depends on three functions: sensing, deciding, and acting. The challenge isn't building each one. It's integrating all three fast enough to be useful.
Sense. Decide. Act. In milliseconds. This is the story of a robot that can beat professional table tennis players. Watch the short film: ace.ai.sony #SonyAI #ACE #Robotics
For 40+ years, building a robot that could rally with an elite human table tennis player at full speed was an unsolved problem. Sony AI's Ace research project set out to change that—and the results are now accepted for publication in @nature.com and featured on the cover.
March at Sony AI: Jesse Lai on writing The Principles of Diffusion Models, 11 papers accepted to ICASSP 2026, and Alice Xiang on the Me, Myself and AI podcast discussing FHIBE. Full recap: bit.ly/4c4oBxO #SonyAI #AIResearch #MachineLearning #ICASSP2026
SHIELD benchmarks how #vision-language models detect #AI-edited images without task-specific training. 👉 Read more: bit.ly/4dRuHE3
EnTruth offers a new way to trace unauthorized dataset use in #diffusionmodels with minimal image alteration. 👉 Full paper: bit.ly/4sXLlpO
StelLA shows how structured low-rank adaptation can outperform standard #LoRA without added compute. 👉 Read the paper: bit.ly/46PkK62
Music Arena brings live, human preference evaluation to text-to-music #research. 👉 Read more: bit.ly/4aMmeQH
FlashFoley enables real-time, controllable sketch-to-audio generation. 👉 Read the paper: bit.ly/46qQQVn
LADiBI solves blind inverse problems using latent diffusion—no training required. 👉 Full paper: bit.ly/46RbrCn
TalkCuts is a large-scale dataset for multi-shot human speech video generation. 👉 Full paper: t.co/AElsaEXUoB
Our 2025 Year in Review is here. This year we advanced responsible data practices with #FHIBE, introduced new tools for music and media creation, strengthened sensing and imaging pipelines, and expanded #RL research from #GTSophy to adaptive agents. Read the full recap: bit.ly/3YG9eoN
LLM-BRec = faster, more personal recs. 50% less training, 80% less inference, better results. 👉 Read the paper: bit.ly/45c7Q0K
