TransFinite AI (@TransFiniteAI) profile picture

Make every model iteration count. https://teapawt.net/t.co/vWrUkfss1c [email protected]

Joined September 2026
Photo by TransFinite AI (@TransFiniteAI)Photo by TransFinite AI (@TransFiniteAI)Photo by TransFinite AI (@TransFiniteAI)
Scaling AI in the physical world requires more than larger models—it demands rich multimodal sensor telemetry and continuous real-world data. From capture to validation, we’re building the dataset for teams building Physical AI. Welcome to TransFinite AI. ℵ
33
Need the same multimodal tactile data? Scaling robot learning with robust tactile data? Volume shipments are live now. Drop us a message to collaborate. huggingface.co/datasets/zhic…
71
If you are at CoRL feel free to send me a DM if you want to chat for physical AI data and more! #CoRL2026
44
TransFinite AI retweeted
GPT and Claude now operate real robot arms, and they generalize remarkably well. What they lack is speed and low-level control. We introduce FACT, the first cerebellum for robots: high-frequency tactile reasoning, learned from scaled human data with both motion priors and applied forces. Brains will keep getting smarter and keep getting replaced, but the cerebellum couples with whichever brain comes next. 1, Built like a brain and a cerebellum. A coarse planner (the brain) takes images, instructions, robot state, past tracks, and predicts future video, tracks and actions. A force-action refiner (the cerebellum) runs fast, jointly predicting fingertip forces and refines actions using new touch feedback. 2, We collect EgoForce, a human manipulation dataset pairing egocentric video and hand motion with calibrated tactile forces. It spans ~2,000 hours, with 40,000 episodes covering 1,000 tasks performed by 1,000 demonstrators. 3, We pretrain FACT, a multimodal robot System 1 policy, on EgoForce plus robot data, then fine-tune with just 10 robot demos + 90 human demos per task. Across 10 real-world tasks, including pipetting, tool use and playing instruments, FACT outperforms π0.5 and T-Rex, with gains up to 60 points over T-Rex. 4, Human force supervision makes a substantial difference. Adding force labels to human pretraining unlocks tasks that were previously almost impossible, such as shoelace tying: Test-tube cap opening: 0% → 80% Flip-cap opening: 10% → 50% Shoelace tying: 0% → 20% FACT connects human contact experience to robot execution-from force supervision during pretraining to predicting and responding to contact on the robot. 📄 Paper: arxiv.org/abs/2610.19155 🌐 Website: fact-1-0.github.io/ Code and dataset coming soon! 🧵 Thread 👇
9
9
70
69,228
TransFinite AI retweeted
Human videos bring scale and motion priors to robot pretraining. But video mainly captures human states, leaving intended actions implicit. Applied forces help fill that gap. Introducing FACT: learning dexterous manipulation from human video and force. 🧵
5
19
128
6,825
Scaling AI in the physical world isn’t just about larger models—it’s about rich multimodal sensor telemetry, and continuous real-world data. From real-world data to validation, we're building infra to make every model iteration count. #PhysicalAI @TransFiniteAI
28
Scaling physical AI requires closing the real2sim2real loop: grounding foundational models in rich multi-modal sensor telemetry, high-fidelity physics simulation, and real-world edge data. Great technical discussions and infrastructure insights here at #TechWeekSingapore. 🇸🇬
30
🚀 Introducing the Next Generation of TouchTronix Robotics FusionX— and our new Vision-Tactile Datasets We’re excited to introduce the latest generation of FusionX, our multimodal data collection system built for robotic learning and Physical AI. This release brings major upgrades across sensing, perception, synchronization, and data annotation: 🔹 State-of-the-art tactile sensors A newly developed thin-film tactile sensor architecture and hot-pressing manufacturing technique delivers significantly improved sensitivity, stability (especially addressing “grasp-free” baseline signal drift), and pixel-to- pixel consistency. 🔹 Three-view visual perception FusionX now combines stereo vision with dual wrist-mounted cameras, capturing manipulation from three complementary perspectives. 🔹 Highly aligned vision-tactile data Tactile signals and visual observations are tightly synchronized and aligned (<5 ms lantency), enabling models to better connect what happens visually with the underlying physical contact. 🔹 In-house hand pose estimation algorithm Our proprietary hand-pose pipeline provides structured hand kinematics alongside tactile and visual observations. 🔹 High-precision natural-language annotations Each dataset can include detailed semantic annotations describing actions, objects, interactions, and manipulation processes — making the data more directly usable for robot learning and multimodal model training. But the most important part is the data. We have already collected thousands of hours of real-world vision-tactile manipulation data, and are scaling toward 10,000+ hours in the coming weeks. Both the complete FusionX system and our vision-tactile datasets are now commercially available. (Product/Datasets contact: [email protected]) We are building the data infrastructure that connects vision, touch, hand motion, gaze, semantics and more— so robots can learn not only what the physical world looks like, but how it feels and how to interact with it. Vision + Touch → Robotic Intelligence. #PhysicalAI #Robotics #EmbodiedAI #RobotLearning #TactileSensing #MultimodalAI #DexterousManipulation #RoboticsData #TouchTronix #FusionX
1
1
10
19,750