Walking Was Solved. Grasping Is the Last Frontier. A Beijing Startup Just Raised $100M+ to Prove It.

The humanoid robots that ran the Beijing half-marathon this year were solving 2015's problem. Walking, balancing, and not falling down: that was the gauntlet. The harder problem is still open. Teaching a robot hand to pick up a wine glass without crushing it, or to fold a spatula around a block of tofu without snapping it in half, is the discipline robotics calls manipulation, and almost no one has shipped it yet.

Four-month-old, Beijing-based PokeBot (破壳机器人) just raised a nine-figure USD Pre-A round to attack that problem. The check, reported by TechTimes and confirmed by Chinese investment trade outlet PE Daily, was co-led by Shunwei Capital (顺为资本) and IDG Capital (经纬创投). Zhongding Capital, Jiukun Ventures, Junshan Capital, SEE Fund, Liepin Investment, and Yuanno Capital participated. Existing shareholders Yunqi Capital, Xiaomi Strategic Investment, Honghui Fund, Inno Angel Fund, and Dongfang Jiafu followed on. The deal makes PokeBot the second nine-figure Chinese humanoid-robotics round in roughly three months, per Gasgoo Auto News, following an earlier angel round.

What "manipulation" actually is

Robotics splits cleanly into two disciplines. Locomotion is about getting a body from A to B without falling over. Manipulation is about a hand or gripper doing something useful once it arrives: picking, placing, twisting, cutting, folding. The first has been a solved engineering problem for years. The second is why your dishwasher still does not unload itself.

NetEase's exclusive on the round frames manipulation as the biggest remaining barrier to embodied-AI commercialization, breaking it into five concrete sub-problems: grasping unfamiliar objects, fine force control, tool use, bimanual coordination, and recovery from failure. PokeBot's technical pitch is a "WAM × RL × DATA" flywheel, a World Action Model trained with reinforcement learning on a continuously refreshed data loop. Coverage attributes the architecture to the company; independent verification of the pipeline has not been published.

The public proof of concept

The demo that anchored the round's coverage: a robot cooking mapo tofu. The dish is a useful benchmark candidate because it requires precise timing, gentle handling of soft ingredients, and controlled force when stirring. That is the kind of messy, contact-rich task that exposes manipulation limits quickly. It is a company-supplied demonstration rather than an independent benchmark, and the round should be read on that basis.

What the cap table signals

Shunwei is Xiaomi founder Lei Jun's venture arm. Xiaomi Strategic's participation as a follow-on is a deliberate signal. IDG is a long-cycle China tech investor that has backed robotics bets before. The roster pairs one strategic who can plug PokeBot into a hardware supply chain and one financial who can underwrite a multi-year research burn. The combination reads as a bet that manipulation, not locomotion, is the next product cycle.

The caveat enterprise buyers should weigh

TechTimes' coverage flags Article 7 of China's National Intelligence Law (2017), the Data Security Law (2021), and the Cybersecurity Law (2017) as a deployment-governance factor. That is the legal framework that obligates Chinese organizations to cooperate with national intelligence work and governs how data generated by domestic systems can cross borders. For enterprise operators evaluating PokeBot or its peers for cross-border deployment, this framework is a vendor-assessment reality, not a marketing footnote.

A note on the round size

Chinese Pre-A rounds described as "hundred-million-dollar-plus" almost always refer to 1亿+ RMB, roughly $14M USD at recent rates, rather than the $100M+ USD figure implied by a literal English reading. The dollar figure above tracks the coverage. The currency basis is genuinely ambiguous and the round's headline number should be read as "nine-figure local-currency" until a regulatory filing or audited disclosure confirms otherwise. The discipline-shift thesis (capital rotating from walking to grasping) holds either way.

The next checkpoint is whether PokeBot can move from a mapo-tofu demo to a repeatable manipulation benchmark that an enterprise procurement officer would sign off on. The investor roster suggests the capital is patient enough to wait. The legal caveat suggests the buyer side will be selective about which customers can deploy.