Kepler Robot: Continuously Optimizing Load Capacity and Operational Endurance to Drive the Integration of the VLA Model with the Industrial World Model

When it comes to stock trading, just rely on the Golden Qilin analyst research reports—authoritative, professional, timely, and comprehensive. Help you uncover high-potential theme opportunities!

（Source: Securities Times）

With a focus on developing blue-collar humanoid robots, Shanghai Kepler Robot Co., Ltd. (hereinafter referred to as “Kepler Robot”) will further push humanoid robots into real-world scenarios this year, such as industry, logistics, and special operations, to achieve on-the-ground capabilities that are mass-producible and capable of continuous, sustainable operation.

Kepler Robot’s Chief Technology Officer, Xie Ao, said in an interview with a reporter from Securities Times that this year, the company will continue to optimize the high-load and long-endurance capabilities of its K2 “Bumblebee” humanoid robot. At the same time, it will work to integrate its VLA model (vision—language—action) with its Industrial World Model to achieve more intelligent and more autonomous operation. In terms of scenario applications, the company will continue to expand K2 “Bumblebee”’s practical application cases, including loading and unloading automotive parts, logistics handling, and autonomous inspection, and will also strengthen its Industrial Brain scheduling system to improve multi-robot coordination efficiency and scenario adaptability.

“ The company is accelerating the development of a body intelligence technology framework centered on real data, and will soon release the country’s first native VTLA (vision—tactile—language—action) full-sensing force-tactile data acquisition solution.” Xie Ao said.

Kepler Robot was established in 2023 and has always focused on building “blue-collar humanoid robots,” enabling application scenarios such as intelligent manufacturing, warehouse and logistics, special operations, and scientific research data collection. Through independently developed key components such as a Xingxing roller screw actuator and a dexterous hand, combined with self-developed control algorithms, the K2 “Bumblebee” robot has achieved ultra-long endurance with “charging for 1 hour and working for 8 hours,” as well as a heavy-load capability of transporting 30 kilograms with dual-arm coordinated handling.

To date, the K2 “Bumblebee” robot has already run through a positive closed loop of “scenario-driven—data accumulation—rapid iteration—scale delivery,” and has also entered sites such as the SAIC-GM logistics factory, the parts production workshop of Zhaofeng Co., Ltd., and the Puremi Technology factory. It has also taken part in the world’s first human-robot collaborative high-altitude welding operation.

According to information, during this year’s Spring Festival, dozens of Kepler K2 “Bumblebee” units became special workers at the Puremi Technology factory. They charge for 1 hour and keep working for 8 hours, with an overall job success rate of about 98%. In terms of specific functions, besides efficiently completing tasks such as picking workpieces, loading and unloading, and moving tote boxes in the warehouse area, K2 “Bumblebee” also handles plant inspections and anomaly monitoring, providing around-the-clock support for logistics and production cycle timing.

“ We first built industrial scenarios within the company, aiming to achieve a one-to-one reconstruction, including machining parts, cutting parts, workbenches, and machine tool models, etc.” Xie Ao said when describing the R&D and iteration process of “blue-collar robots.” “ Then, through large-scale data acquisition using methods such as exoskeletons and VR, we pre-train industrial VLA models. Next, we will carry out small-scale data acquisition in actual industrial sites, and then perform model fine-tuning and optimization.”

Xie Ao said that, around the R&D of “industrial blue-collar robots,” the company still has several development bottlenecks that urgently need to be overcome, including environmental perception and dynamic planning capabilities in highly complex scenarios, reliability and safety of robots during long-cycle, high-intensity operations, and the integration efficiency and stability between large-scale world models and VLA models, among others.

Against the backdrop of a new round of technological revolution and industrial transformation accelerating, artificial intelligence is moving from competition in “algorithms and compute power” to a new stage of “deep integration of intelligent agents and the real world.” Body intelligence robots are an important carrier of the deep coupling between AI and the physical world. In the industry, it is widely believed that world models enable AI to truly understand and interact with the physical world, and that this is the necessary path for body intelligence robots to move toward AGI (artificial general intelligence).

Xie Ao told reporters that Kepler Robot has already begun building an Industrial World Model and a Home World Model. Next, the plan is to first integrate the Industrial World Model with the Industrial VLA model, and verify its effectiveness through small-scale POC (feasibility validation) to lay a foundation for future large-scale deployment.

“ From world model research and iteration to the final commercialization product, there is still a long road ahead.” Xie Ao said. Even if the development speed of the world model is temporarily not up to expectations, there are many feasible compromise technical solutions. Although these solutions may not be perfect yet, they can lay a solid foundation for the final deployment of world models and promote the gradual commercialization of body intelligence in real-world scenarios.

It was also reported that Kepler Robot will release soon the country’s first native VTLA (vision—tactile—language—action) full-sensing force-tactile data acquisition solution. This solution builds a high-fidelity, multi-modal aligned data closed loop in real industrial scenarios, addresses the shortcomings of traditional VLA models in the force-tactile and physical interaction layers, provides key data support for the deep integration of the Industrial World Model and sensing decision-making capabilities, and further explores how to implement a data-driven approach to reconstructing industrial intelligence.

Massive information, precise interpretation—exclusively on the Sina Finance APP