TESOLLO Showcases VLA and Dexterous Manipulation at ICRA 2026: From Teleoperation to Humanoid Robotics
- 2 days ago
- 5 min read
As humanoid robotics continues to advance, one challenge remains particularly difficult: enabling robots to manipulate objects with the same dexterity as humans. While significant progress has been made in locomotion, navigation, and environmental perception, precise hand manipulation is still one of the most critical capabilities for real-world deployment.
At ICRA 2026 (IEEE International Conference on Robotics and Automation) in Vienna, TESOLLO presented a series of technologies centered on dexterous manipulation, including teleoperation, in-hand manipulation, humanoid robot integration, and demonstrations powered by Vision-Language-Action (VLA) models.
Together, these demonstrations highlighted not only the capabilities of TESOLLO's dexterous hand platform, but also several emerging trends shaping the future of Embodied AI, Physical AI, and humanoid robotics.
AXIS 3D Insights: Dexterous Manipulation Is Becoming a Critical Piece of Embodied AI
Our team attended ICRA 2026 in Vienna to experience these demonstrations firsthand.
One clear takeaway from this year's exhibition was that regardless of whether the focus is teleoperation, Vision-Language-Action models, or humanoid robots, every intelligent system must ultimately prove one thing—its ability to perform reliable physical manipulation.
TESOLLO's demonstrations reflected a broader industry direction. The next generation of AI is no longer defined solely by its ability to understand the world, but by its ability to interact with it through precise, physical actions.
Exhibit 1: Dexterous Hands Are Becoming the Core of Robotic Manipulation
TESOLLO's showcase featured two flagship five-finger dexterous hands: the DG-5F-M and the DG-5F-S.
The DG-5F-M offers 20 independent degrees of freedom (20 DoF), enabling complex grasping strategies, tool use, and advanced manipulation research. Designed for research environments, it provides the flexibility required for developing sophisticated robotic manipulation algorithms.
The DG-5F-S, meanwhile, adopts the same five-finger architecture while introducing a lighter and more compact design optimized for humanoid robot integration. Its smaller form factor allows developers to deploy dexterous manipulation capabilities across a wider range of humanoid platforms.
One of the most noticeable shifts throughout the exhibition was the industry's changing focus. The conversation is no longer about whether a robot can simply grasp an object—it is increasingly about whether it can manipulate that object with human-like precision.
This direction aligns closely with the development strategies pursued by leading humanoid robotics companies such as Figure, Agility Robotics, and Tesla Optimus, all of which view dexterous manipulation as a fundamental capability for future autonomous robots.
Exhibit 2: Teleoperation Is Becoming a Key Source of Robot Training Data
Beyond hardware demonstrations, TESOLLO also showcased applications based on teleoperation.
The value of teleoperation goes beyond enabling humans to directly control robots in real time. More importantly, it provides an efficient way to collect high-quality human demonstration data, which can later be used for Imitation Learning and the training of Robot Foundation Models.
In recent years, many Embodied AI research teams have begun developing workflows based on:
Human Operation → Data Collection → AI Learning → Autonomous Execution
Within this pipeline, teleoperation has become an essential component for bridging human skills and robotic intelligence.
From this perspective, TESOLLO's demonstration was not only about showcasing a dexterous hand platform. It also represented an important data acquisition interface within the broader robot learning ecosystem.
Exhibit 3: VLA Is Moving Beyond the Laboratory
Another notable highlight from the exhibition was the demonstration of humanoid robot manipulation powered by Vision-Language-Action (VLA) models.
Traditional large language models primarily focus on understanding text and knowledge. VLA models extend this capability by integrating three key elements into a unified framework:
Vision — Understanding visual information from the environment
Language — Interpreting human instructions and intentions
Action — Executing physical movements through robotic systems
This enables robots to understand natural language commands and perform corresponding tasks in real-world environments.
TESOLLO's demonstration highlighted how dexterous hands are becoming a critical execution component for bringing VLA systems into practical applications.
When an AI model understands an instruction such as "pick up a cup," the robot still needs a high-degree-of-freedom robotic hand to physically complete the task. This is why dexterous hands and VLA models are increasingly discussed together as key technologies for the future of embodied intelligence.
Exhibit 4: DG-5F-S Reflects the Commercialization Needs of Humanoid Robotics
While the DG-5F-M represents advanced manipulation capabilities for research and complex applications, the DG-5F-S reflects another important direction in the humanoid robotics industry: practical deployment.
The DG-5F-S maintains a five-finger, 20 DoF architecture, while adopting a more compact and lightweight design to improve integration flexibility and reduce implementation barriers. TESOLLO positions the DG-5F-S as a dexterous hand solution designed for humanoid robot platforms and real-world deployment, rather than being limited to laboratory research.
From an industry perspective, this represents a significant shift:
Humanoid robots are moving from research environments toward commercial applications.
As the market evolves, dexterous hand development is also shifting from pursuing maximum performance alone toward achieving a balance between capability, cost, and system integration.
Three Key Trends Observed at ICRA 2026
Through TESOLLO's demonstrations at ICRA 2026, three major trends shaping the future of robotics become increasingly clear:
1. Dexterous Manipulation Is Becoming a Core Capability of Humanoid Robots
Robots are no longer expected to simply move through environments. Future humanoid robots must be capable of performing complex manipulation tasks in real-world scenarios, from handling objects to using tools with human-like precision.
2. Teleoperation and Data Collection Are Equally Important
High-quality human demonstration data is becoming a critical foundation for training Embodied AI models.
As robots move toward greater autonomy, collecting real-world interaction data through teleoperation will play an increasingly important role in developing intelligent robotic systems.
3. VLA and Physical AI Are Accelerating Real-World Deployment
AI is moving beyond simply understanding the world. Through robotic systems, AI is beginning to interact with and influence the physical environment.
The combination of Vision-Language-Action models, dexterous manipulation, and robotic platforms represents a major step toward practical Physical AI applications.
Conclusion
TESOLLO's demonstrations at ICRA 2026 reflect the broader technological evolution taking place across the humanoid robotics industry.
From high-degree-of-freedom dexterous hands and teleoperation systems to in-hand manipulation and VLA-based robotic control, the industry is gradually moving beyond perception and decision-making toward a new era of Embodied AI with real-world physical capabilities.
Dexterous hands, once primarily viewed as research platforms, are now becoming a critical interface connecting artificial intelligence with the physical world.
As humanoid robots continue to advance, the ability to manipulate, learn, and interact with their environment will become one of the defining factors determining how quickly intelligent robots transition from laboratories into real-world applications.
About AXIS 3D
AXIS 3D is the official TESOLLO distributor in Taiwan, specializing in motion capture, XR, virtual production, and AI robotics integration.
We help organizations turn emerging technologies into practical solutions.
