Nvidia's First Global Haptic Simulation Partner Secures Hundreds of Millions in Financing, H1 Orders Quadruple Last Year
Beijing Tashan Technology completed a B-round financing of hundreds of millions of RMB, led by industrial investors including Joyson Electronics and Aux, to accelerate haptic sensor and chip iteration. The company holds over 80% market share in humanoid robot haptic sensors, with order volumes in the first half of 2026 exceeding four times the previous year's full-year revenue. Tashan has developed a full-stack technology system covering custom AI haptic chips based on spiking neural networks, m
Analysis
TL;DR
- Beijing Tashan Technology completed a B-round financing of hundreds of millions of RMB, led by industrial investors including Joyson Electronics and Aux, to accelerate haptic sensor and chip iteration.
- The company holds over 80% market share in humanoid robot haptic sensors, with order volumes in the first half of 2026 exceeding four times the previous year's full-year revenue.
- Tashan has developed a full-stack technology system covering custom AI haptic chips based on spiking neural networks, multi-modal sensors, and simulation platforms integrated with NVIDIA Isaac Sim.
- Haptic perception is rapidly transitioning from an optional feature to a standard requirement in dexterous hands, driven by the need for closed-loop physical interaction in embodied AI.
- Strategic partnerships include collaborations with Turing Award winner Richard Sutton for a "Robot Kindergarten" and NVIDIA for global haptic simulation, aiming to solve data scarcity in tactile learning.
Why It Matters
This development highlights the critical shift in embodied AI from visual-only perception to multi-modal sensory integration, where touch is becoming the bottleneck for precise manipulation and safe human-robot interaction. For AI practitioners and hardware manufacturers, the rapid adoption of haptic sensors indicates that future robustness in physical tasks depends heavily on high-fidelity tactile data and specialized edge-processing chips. The industry-wide consensus on tactile necessity suggests that companies lacking haptic capabilities may face significant limitations in achieving general-purpose robotic dexterity.
Technical Details
- Custom AI Chip: Developed the world's first mixed-signal AI haptic chip based on Spiking Neural Networks (SNN), enabling low-latency, low-power edge processing of multi-dimensional tactile signals. A next-generation large chip has been taped out and is scheduled for release in Q3 2026.
- Multi-Modal Sensors: Core products include the TS-F fingertip sensor (integrating proximity, 3D force detection, temperature, and texture recognition with 0.01N resolution) and the TS-E mechanical hand sensor. Measurement frequency has increased 3-4 times year-over-year, with upcoming models adding temperature sensing.
- Simulation and Data Infrastructure: As NVIDIA’s first global haptic simulation partner, Tashan has open-sourced its haptic simulation to MuJoCo and NVIDIA platforms. They have established data collection centers in Beijing and Hubei focusing on heterogeneous body adaptation and autonomous data acquisition.
- Algorithm Platforms: Launched the TS-V vision-haptic fusion training platform and TS-VT data collection version to lower the barrier for tactile algorithm training. Collaborated with Richard Sutton to explore autonomous trial-and-error learning in real physical environments.
Industry Insight
- Supply Chain Consolidation: With Tashan dominating over 80% of the niche market, downstream robot manufacturers are increasingly dependent on this single supplier for core haptic components, creating potential supply chain risks but also validating the commercial viability of the technology.
- Standardization of Touch: The transition of haptic sensors from "optional" to "standard" equipment in dexterous hands suggests that future robotics benchmarks will likely include tactile performance metrics, driving demand for standardized interfaces and data formats.
- Data-Centric AI Evolution: The emphasis on building dedicated haptic data centers and simulation platforms indicates a strategic pivot towards data-centric AI, where high-quality, multi-modal tactile datasets will become as valuable as visual datasets for training embodied intelligence models.
Disclaimer: The above content is generated by AI and is for reference only.