DexForce Teaches Humanoids Real-World Skills

DexForce, a robotics firm in Shenzhen, China, is virtually training its humanoid robots to help reduce cost and time.  

The company has developed AI software that puts a digital spin on the workplace. In this way, robots can get an understanding of how objects will act and practice on their movements before applying those skills to physical machines. 

As Reuters reported, it is a means of surmounting an industry challenge: amassing the sort of training data that makes a robot more than just a controlled demo piece. 

Training Robots in Virtual Workplaces 

At DexForce's headquarters, virtual robots practise tasks such as moving objects across a table. The simulations are run through various changes of position and circumstance so the AI is put to the test in different scenarios, all without the need for a physical demonstration. When the training is done, the software transfers the learned behavior to real robots. 

DexForce maintains that it lessens the need for human trainers; they might have to go through the motions 100 times or more to generate enough data in front of the robot. That being said, the developers can put together scenarios in a virtual space that would be hard or costly to reproduce physically. 

Transferring skills from simulation to reality, though, is something that remains challenging. A robot has to contend with the vagaries of lighting, surfaces and obstacles it has not come across before. 

$147 Million Funding Backs Expansion 

Jia Kui, a machine-learning professor at the Chinese University of Hong Kong, started DexForce in 2021. The business has some 500 staff now and has been valued at more than 10 billion yuan after a Series B funding of 1 billion yuan ($147 million). Its investors include Lenovo Capital, Lens Technology and a number of Chinese state-backed funds. 

Bloomberg reports that an IPO in Hong Kong is under consideration by DexForce, but Jia declined to comment, according to Reuters. It is an investment that speaks to the commercial appeal of AI systems for the control of working robots. 

From Simulations to Real-World Applications 

Reuters reports manufacturers from Toyota to Samsung, Panasonic, Foxconn, Haier and Hisense have made use of the software in industrial settings. On the consumer side, the company’s humanoids are to be found in over 15 cities in China performing tasks, such as make coffee for visitors at tourist spots and shopping centres.  

Detailed figures on robot autonomy, operating costs and long-term reliability remain limited, but these applications demonstrate how the company's technology is being introduced into commercial environments. 

What remains to be seen is whether DexForce can show that skills developed in a virtual space can consistently translate into productive work. Should it, the method may well be one for other robotics firms to follow in order to build capabilities without the tedium of gathering extensive physical data.