Just Over a Year Since Inception, LumiBot Challenges Embodied AI's Data Bottleneck at IROS 2026

PITTSBURGH, Sept. 29, 2026 /PRNewswire/ -- On the exhibition floor of IROS 2026, LumiBot (known in China as Shiyue Technology)—a startup founded just over a year ago—is showcasing its latest finger-shaped visuo-tactile sensors, dexterous hand series, and manipulation foundation models. Beyond its exhibition booth, LumiBot has also been selected as a high-potential startup to present at the official IROS Startup Session.

At the heart of LumiBot's showcase is a core thesis: the upper bound of data quality is strictly determined by hardware reliability. In embodied AI, high-quality real-world manipulation data remains critically scarce. Industry-wide, public multimodal datasets featuring high-DOF dexterous hands with tactile feedback total only a few tens of thousands of samples, whereas training scalable manipulation models requires tens of millions of hours of real-world operational data. LumiBot tackles this bottleneck at the hardware source—utilizing highly reliable, high-precision hands and visuo-tactile sensors to ensure that collected data is standardized, clean, and reusable from day one.

High-Reliability Hardware Guarantees Continuous and Stable Data

The Lumi-Dex hand series supports mainstream industrial communication protocols and seamlessly integrates across various robotic arm platforms and simulation environments, serving as a unified execution endpoint for diverse embodied architectures. On-site, LumiBot is dynamically demonstrating its high-DOF hands, notably the Lumi-Dex-21, which features an innovative drive-control integrated design combining linkage and direct-drive mechanisms.

Furthermore, LumiBot's upcoming flagship dexterous hand brings substantial upgrades in lifespan, durability, and thermal management. Powered by a self-developed active-cooling micro joint module, it maintains operating temperatures around 50°C and supports over 5,000 hours of continuous no-load operation, capable of supporting weeks of uninterrupted data collection.

High-Precision Perception Ensures Crisp Data at the Source

The Lumi-Tac visuo-tactile sensor V2 incorporates structural and optical optimizations over its predecessor, extending coverage across the fingertip, finger side, and finger pad with significantly enhanced wear resistance and deformation reliability. Key specs include a 10-micron spatial resolution, 0.02 N force detection accuracy, and a 120 fps frame rate. Paired with proprietary high-precision curved surface reconstruction algorithms, Lumi-Tac delivers comprehensive and dependable multimodal tactile capture, providing high-information-density data streams for model training.

Closed-Loop Model & Data Alignment from the Source

On the data front, LumiBot has fully operationalized its real-robot teleoperation collection pipeline for continuous data gathering. Simultaneously, its egocentric (Ego) data-to-robot mapping scheme has achieved millimeter-level precision verification, providing stable training inputs.

On the model front, the Lumi-Mind manipulation foundation model utilizes a hierarchical MoT (Mixture of Transforms) architecture that deeply fuses Vision, Tactile, Language, and Action (VTLA) modalities. Through a unique memory mechanism and tactile gating, Lumi-Mind enables long-horizon task planning and fine-grained micro-adjustments during dense contact. Via its proprietary Lumi-Cap data pipeline, LumiBot seamlessly connects Ego video capture, teleoperation data gathering, and automated data cleaning into a robust "capture-train-deploy-feedback" flywheel, aligning data with model training requirements right from the start.

Live Demo: Completing Long-Horizon Tasks with a Constrained-DOF Hand

At IROS, LumiBot demonstrated its self-developed Lumi-Dex-11 (a lower-DOF hand) driven by the Lumi-Mind model as it executed a complex sequence: "picking up a stamp → precise stamping → picking up a tray → handing out a card → returning the tray → returning the stamp." When human disturbances were introduced, the model autonomously re-evaluated and adjusted its task planning in real time. This demonstration highlights how deep integration of algorithms and hardware enables reliable execution of complete, long-horizon tasks even on hardware with constrained degrees of freedom.

Scaling Up Dexterous Hands for Real-World Deployment

Founded in 2025, the company—just over a year old—has developed its entire product portfolio in-house, spanning motors, visuo-tactile sensors, complete dexterous hands, manipulation models, and data acquisition pipelines. This full-stack self-reliance ensures that collected data is naturally aligned with model training needs, standardized, and reusable—an advantage neither pure hardware nor pure algorithm companies can achieve alone.

The upper limit of data quality is dictated by hardware reliability; however, enabling large-scale acquisition of high-quality data requires that dexterous hands themselves achieve scalable deployment. LumiBot's ultimate goal is to truly scale up dexterous hands, transitioning standardized data collection from research laboratories into real-world industrial environments.

Currently, LumiBot is advancing scenario validation and technical verification (POCs) with industry leaders across automotive manufacturing, 3C electronics, and life sciences, systematically proving the scale-up capability and reliability of its hardware and systems in actual production environments.

About LumiBot
Founded in 2025, the company specializes in robotic dexterous manipulation intelligence. Driven by a core technology stack centered on a deeply integrated 'hardware + data + model' closed loop, it is committed to empowering robots with evolvable and generalizable fine-manipulation capabilities, enabling them to efficiently adapt to complex, dynamic unstructured environments and real-world application scenarios.

Media Contact
Official Website: http://lumirobot.net 
Linkedin: https://www.linkedin.com/company/lumibot-shiyue-tech
Email: [email protected]  

SOURCE LumiBot