Embodied intelligence company Spirit AI has completed an extension to its angel funding round, according to business registration records and insider sources. Broadvision Capital, founded by Li Ping, co-founder and vice chairman of Contemporary Amperex Technology (CATL), led the investment.
Founded by former Rokae Robotics CTO Han Fengtao in February 2024, Spirit AI has raised funds three times within its first nine months. Insiders told 36Kr that the company is already working to secure further funding.
The automotive industry has emerged as a crucial proving ground for embodied intelligence robots, bringing lab innovation onto the factory floor. The industry’s supply chains are notoriously labor-intensive. For instance, CATL employs tens of thousands globally, as noted in its annual report, with a substantial share based overseas.
With an aging population and fewer young workers entering manufacturing, labor costs are rising fast. This trend is also playing out internationally, where Chinese automakers expanding abroad face high labor costs and localization challenges. Humanoid robots may hold the key to these complex issues.
Since early 2024, many embodied intelligence firms have turned to automotive factories for testing. For instance, Figure AI has deployed its latest Figure 02 model at BMW’s Spartanburg plant, and Elon Musk has integrated Optimus robots into Tesla production lines for battery handling, with plans to deploy 100 robots next year.
Automakers’ enthusiasm for embodied intelligence is soaring. According to 36Kr, companies like Tesla, Toyota, BAIC, SAIC Motor, Xpeng Motors, BYD, and Li Auto are stepping into the field, either by developing their own robots or investing in promising players in the industry.
Yet, bringing humanoid robots into automotive production is no simple feat.
On the software side, robots must reach an exceptionally high degree of precision to operate effectively in automotive environments. Spirit AI addresses this with a reinforcement learning framework, which enables its robots to improve their accuracy through real-world practice, helping meet the rigorous standards that factory environments demand.
Cost-efficient data collection is another challenge for these robots. Spirit AI has made strides in high-sample-efficiency imitation learning. By leveraging the SGRv2 framework, Spirit AI’s robots can integrate scene geometry with object semantics to perform behavior training at a data efficiency more than 20 times that of traditional algorithms.
High costs have traditionally hindered humanoid robot deployment in automotive production lines. Currently, a fully functional humanoid robot can cost over USD 1 million, sometimes reaching higher seven-figure sums.
Cost control, however, is where Spirit AI claims to stand out. With extensive experience deploying tens of thousands of commercial robots, the team said it has developed efficient design and manufacturing processes that meet industrial standards without exorbitant costs.
Sources at 36Kr said that Spirit AI’s solution is adaptable to various hardware platforms, enabling a single model to work across different embodied forms. This “one brain, multiple forms” approach significantly lowers costs in automotive production and allows for more tailored, flexible applications in factories.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This article was written by Qiu Xiaofen for 36Kr.