Hand someone an object, wipe a surface, or guide a mechanical part into place, and you are constantly — and unconsciously — adjusting your grip. You feel resistance, ease your pressure, and change direction on the fly.
For modern robots, that seemingly ordinary interaction remains a monumental engineering hurdle.
While AI models have made massive strides in “thinking,” physical execution often falls short because robots frequently lack a nuanced sense of touch. That is the challenge behind GENTO’s proprietary 21-degree-of-freedom dexterous hand, making its global debut at CoRL 2026 (Conference on Robot Learning), running November 9–12 in Austin, Texas. At Booth 602 (Level 3 of the JW Marriott Austin), Tianji Intelligence is presenting the GENTO hand alongside its TIANJI force-controlled arms, TIANJI Marvin Pro platform, and GENTO Luna, extending its approach to robotic manipulation from arms to fingertips. The exhibition area is open until 1:00 p.m. local time on November 11.
One Mechanical Design, Four Persistent Problems
GENTO’s proprietary 21-degree-of-freedom dexterous hand prototype makes its first public appearance worldwide at CoRL. The robots used in GENTO’s on-site demonstrations are also equipped with the brand’s own dexterous hands. Its core proposition is straightforward: one linkage-driven mechanical design that addresses four longstanding problems together — high backlash, fragile transmission mechanisms, joints that cannot be backdriven, and limited rated fingertip force for a given finger size.
First, high-precision, low-clearance linkage transmission reduces backlash — the mechanical play that can make finger movements less consistent. Second, a transmission structure designed to withstand impacts and overloads addresses the fragility that can turn an accidental collision into a repair. Third, every joint is backdrivable, allowing an external force to move the fingers rather than meeting rigid resistance. Paired with MIT force/position hybrid control, this supports more compliant physical interaction. Fourth, efficient force transmission through the linkage mechanism delivers greater rated fingertip force within the same finger dimensions.
GENTO’s emphasis is on how these capabilities work together within one architecture. The design tackles precision, durability, backdrivability, and fingertip force simultaneously, addressing the mechanical limitations that have long made dexterous hands difficult to use reliably.
Why Touch Is the Missing Link for Embodied AI
The wider robotics field includes NVIDIA’s Isaac development platform and Tesla’s Optimus humanoid robot. A robot’s movement can look flawless in a simulation and still fail miserably in the real world. During precision assembly, a fraction-of-a-millimeter misalignment can spike the resistance a robot encounters. When wiping a table, success isn’t about moving to a coordinate; it’s about maintaining continuous, adaptive contact pressure.
Tianji Intelligence’s approach tackles this by combining torque sensing with compliant impedance control. In plain English: the robot measures twisting forces dynamically at every joint and adjusts its stiffness in real time. It can lock down with rigidity when a heavy task demands it, or instantly yield when an interaction calls for a soft, human-safe response — fulfilling GENTO’s brand promise of “Touching the World Gently.”
For AI researchers, this solves a critical data collection bottleneck. Standard cameras capture what moved, but force data records how the physical interaction felt. By feeding force-coupled data into training pipelines, Tianji Intelligence aims to bridge the gap between digital AI policies and messy real-world physics.
Inside the Hardware: High-Resolution MEMS and Benchmarked Joints
Tianji Intelligence’s hardware ecosystem traces back to its industrial roots — having been jointly incubated by industrial robotics powerhouse Yaskawa Electric and precision manufacturer Everwin Precision, with over 30,000 industrial robots deployed globally.
The foundation of its current lineup relies on in-house developed MEMS joint torque sensors — a technology adapted from consumer electronics to robotics — engineered down to detect torques as minuscule as 0.02 N·m with a 160 Hz torque-loop bandwidth and repeat positioning accuracy of ±0.03 mm. Integrated into the TIANJI-branded robotic arm lineup (M6-Lite, M6, and the heavy-duty M20 offering up to 20 kg payload per arm), these sensors ensure every joint acts as an active sensory organ without needing frequent recalibration.
Furthermore, Tianji Intelligence’s integrated joint modules pack the motor, reducer, driver, encoder, and brake into a single unit that is roughly 25% lighter than key competitors. To back up its claims, Tianji Intelligence released open trajectory tracking benchmark scripts on GitHub (May 2026), demonstrating tracking error reductions from 10.56 mm down to 1.89 mm and latency dropping from 45 ms to 5 ms for third-party verification.
Supporting Platforms: GENTO Luna and TIANJI Marvin Pro
Moving from fixed arms to mobile platforms, GENTO extends whole-body force control to dynamic environments.
- GENTO Luna: A wheeled, foldable humanoid platform featuring an omnidirectional mobile base, dual arms, and a torso. At CoRL 2026, GENTO is running a teleoperation-driven demonstration where Luna, equipped with GENTO’s in-house dexterous hands, mirrors visitor movements and handles item handovers in real time.
- GENTO Skye: A lift-column wheeled configuration tailored specifically as a robust development platform for reinforcement learning and real-world deployment validation.
Alongside Luna is TIANJI Marvin Pro, a fixed-base dual-arm platform designed for teleoperation and real-robot data collection. At Booth 602, visitors can test a voice-command grasping demo where TIANJI Marvin Pro parses natural language instructions to identify and retrieve specific objects from a group.
NVIDIA Launches GR00T 2 With TIANJI Marvin Pro
At CoRL 2026, NVIDIA introduced GR00T 2, an open-world action model built on NVIDIA Cosmos that predicts future video and robot actions together. The launch demonstrations featured a TIANJI Marvin Pro platform, developed by Tianji Intelligence for NVIDIA with a tailored head and body.
Equipped with two dexterous hands, the platform performed tasks requiring coordinated manipulation. In one skill-transfer demonstration, researchers recorded human hand movements with a wearable exoskeleton, then used GR00T 2 to carry out tasks such as preparing banana yogurt on the robot. This demonstrates TIANJI Marvin Pro’s role as a physical platform for executing learned robot behaviors.
GENTO robots have arrived at the APEC venue and will provide services during the upcoming meetings. CoRL 2026 photos show TIANJI Marvin Pro and the 21-DoF dexterous hand prototype alongside the APEC venue photo.
From Human Demonstration to AI Data Pipelines
Hardware is only half the equation; data is where robot learning thrives. Tianji Intelligence has built a comprehensive data workflow connecting VR headsets, data gloves, and leader-follower dual-arm teleoperation with an annotation and format-conversion platform.
When a human operator guides a robot through a contact-heavy task, the system logs exact resistance alongside synchronized vision and motion. Tianji Intelligence estimates that a fleet of 100 TIANJI Marvin Pro platforms can collect over 600,000 real-robot demonstration datasets annually. This rich data pipeline also serves as the physical testing bed for cutting-edge AI models, such as Shengshu Tech’s Motus2 world model.
The Hidden Infrastructure Scaling Embodied AI
While headlines focus on ‘who is building humanoids,’ Tianji Intelligence is quietly cementing its position as the underlying execution layer for the industry. In mid-2025 alone, the company delivered over 2,000 force-controlled humanoid dual arms within four months, holding the top market share in force-controlled arms.
Tianji Intelligence now powers more than 45 robotics and embodied-AI partners globally. Its client and partner ecosystem includes robotics developers, embodied-AI companies, and established technology manufacturers. Backing this rapid expansion is substantial institutional funding, including a 1 billion RMB Series B/B+ round led by Hillhouse Venture Capital and Meituan, followed by a B++ strategic round led by Ant Group and Sofina.
If you are attending CoRL 2026 in Austin, experience the hardware firsthand at Booth 602 (Level 3, JW Marriott Austin). The exhibition area is open until 1:00 p.m. local time on November 11. Media representatives can schedule a dedicated walkthrough or request the full technical spec sheet by contacting Global PR Manager Eva Tao ([email protected]) or visiting www.gentorobotics.ai.
About GENTO & Tianji Intelligence
GENTO is the embodied robotics brand under Tianji Intelligence. Its portfolio spans dexterous hands and humanoid robots including GENTO Luna and GENTO Skye. Robotic arms and the TIANJI Marvin Pro teleoperation and data-collection platform are marketed under the TIANJI brand. Tianji Intelligence was jointly incubated by Yaskawa Electric and Everwin Precision. With over 30,000 industrial robots delivered globally and over 2,000 force-controlled arms deployed in 2025 alone, Tianji Intelligence serves a broad range of robotics and technology customers. Learn more at www.gentorobotics.ai.