2026-11-13 10:00

Introduction: Robots adapt to new scenarios through demonstration without modifying model parameters. Four months into entrepreneurship, Li Jianxiong is still adjusting to the rhythm of startup life. Information on fundraising, technology, organization, and collaboration floods in simultaneously, leaving this 27-year-old founder and CEO of BASAL Intelligence feeling his "contextual input has exploded and even overloaded."

Four months into entrepreneurship, Li Jianxiong is still adapting to the post-startup pace.

Rapid influxes of information on fundraising, technology, organization, and collaboration leave this 27-year-old founder and CEO of BASAL Intelligence feeling his "contextual input has exploded and even overloaded."

BASAL Intelligence was founded in July 2026, a company specializing in embodied intelligence. The core team consists of seven researchers who previously collaborated long-term at the Institute for AI Industry Research (AIR) at Tsinghua University, with Dr. Zhan Xianyuan, Deputy Researcher at AIR, serving as Chief Scientist. On October 8, the company completed its seed round, led by WuYuan Capital, with follow-on investments from Ivy Capital, Linear Capital, and Huashan Capital.

Light Source Capital serves as BASAL Intelligence’s exclusive financial advisor. Recently, its Managing Director Li Zhengwei frequently discussed with Li Jianxiong the question of how to allocate funds raised. “For a young team just starting out, this is a substantial sum,” Li Zhengwei said.

In the first half of the year, as Li Jianxiong approached graduation from Tsinghua AIR’s doctoral program, he received offers from major tech companies, including one with an annual salary close to eight figures. After brief hesitation, he ultimately chose entrepreneurship.

Photography: She Guisen

BASAL Intelligence does not develop robot hardware but instead focuses on action modeling, pursuing the In-context learning paradigm—enabling robots to acquire new capabilities without retraining model parameters each time, but rather by adjusting context to adapt to new environments and tasks.

"We hope that in the future, robots won’t be able to do everything right away, but instead possess human-like abilities to learn quickly and adapt swiftly. Only then can robots truly enter diverse real-world scenarios," Li Jianxiong said. In essence, he envisions robots capable of gradual self-evolution, rapidly adapting to new environments and tasks, thus enabling large-scale deployment.

This vision aligns with his research at AIR, yet he realized the resources required to achieve it far exceed what academic research alone can provide. "Whether from necessity or resource requirements, this must be accomplished within a corporate structure."

On one hand, a technology path requiring long-term investment; on the other, the immediate challenges startups face—fundraising, organizational building, and commercialization. Li Jianxiong describes entrepreneurship as "living day by day, dying day by day." He told China Entrepreneur that this decision represents the riskiest and most impulsive choice he’s ever made in his life.

Feeling Stuck

April 2026, the deadline for two job offers loomed.

One offered a total package close to eight figures—so high it felt like astronomical numbers; the other came from a major internet company, with comparable compensation but work more aligned with his interests. Li Jianxiong chose the latter.

But after joining the big tech firm, he found he couldn’t fully commit to the work he truly cared about. In large corporations, everyone specializes in a single module, while embodied intelligence requires a full-stack, end-to-end industrial chain. Once his component was completed, if he identified issues in other modules, he lacked the authority to drive improvements—leading to intense frustration.

"Every day I spent there, I felt trapped by uncontrollable factors," Li Jianxiong said.

He began realizing only entrepreneurship could allow him to fully implement his vision of a technical system based on his own understanding. As graduation approached, this feeling intensified. Yet he sensed that delaying entrepreneurship would mean wasting time and life.

"Facing a high-paying offer, 'normal' people might first think: work for two years, earn a few million, then start a company later—with big tech experience as a credential. Isn’t that better? Why take such a 'stupid' risk as full-time entrepreneurship?" Li Zhengwei said. But after interacting with Li Jianxiong, he realized Li wasn’t a 'normal' person by conventional standards. "He’s someone with a strong gambling instinct—he’s betting on his ceiling."

After deciding to launch, he convened a meeting with six former lab colleagues. They instantly aligned.

The team’s exploration of In-context learning began in the summer of 2024, when the VLA concept was emerging. After two months, they voluntarily paused the project. Li Jianxiong admitted they had underestimated the implementation difficulty—many prerequisite modules were still immature. Continuing would have resulted only in small-scale academic proof-of-concept demonstrations, not a viable system.

The team then broke down the problem and prioritized filling missing components. In November 2024, they developed UniAct—a universal cross-body action encoder. Thereafter, they continued research along this path until conditions matured enough to resume focus on In-context learning.

This approach doesn’t aim for robots to master everything immediately, but rather equips them with rapid learning capabilities.

"Though small in size, our team is fully equipped. Fundamentally, we are a full-stack robotics team with hard-earned accolades—Global Champion at the Zhiyuan Robotics Challenge 2025 (IROS), Outstanding Paper Award at ICML 2024 MFM-EAI Workshop, Best Paper Award at ICLR 2026 Efficient Spatial Reasoning Workshop. Whether in academia or industry, these prove our team’s strength. Starting a company is a natural progression," Li Jianxiong said.

WuYuan Capital led the seed round investment in BASAL Intelligence. During due diligence, investors experienced firsthand the team's cohesion. Frequently, nearly ten team members sat across from them, including Professor Zhan Xianyuan. When asked questions, the group engaged in real-time discussion rather than following pre-assigned roles. Members shared deep synergy from joint work and complemented each other effectively around complex problems.

According to the team’s vision, robots should replicate actions through human demonstration, correct errors autonomously, and generalize capabilities across different physical embodiments. When adapting to new tasks, they wouldn’t need to update model parameters per task, but instead adjust behavior via contextual inputs such as demonstration and feedback.

Li Jianxiong hopes this reduces deployment costs. If every new environment demands fresh data collection, retraining, and debugging, costs escalate with scale. Only by accelerating adaptation can costs eventually decline.

However, these capabilities remain current R&D goals. BASAL Intelligence still needs to demonstrate whether its models can operate stably in open environments, whether true cross-body and cross-scenario generalization is achievable, and whether these capabilities can evolve into sustainable products.

Riding the Pirate Ship

During the first month of entrepreneurship, Li Jianxiong’s primary focus was fundraising.

This became the aspect he struggled with most. His first investor meeting involved being handed a piece of paper and a pen, asked to explain his vision in under fifteen minutes. After finishing, he was drenched in sweat—not from heat, but nerves.

He noticed a mismatch in communication style. Previously, both he and his team were overly conservative and reserved. "We achieved 10 points, but might only convey 5."

One investor observed this isn’t age-related; founders without capital market experience or fundraising history typically struggle similarly. "Fundraising is doing a 60-point job, telling an 80-point story, and blowing 100-point hype."

Others told Li Jianxiong that fundraising is like multiplying your value by 1.5—when you’re truly 100 points, you must present as 150. "But currently, we haven’t even reached 1.5—we’re still at 1.0," Li Jianxiong said.

Light Source Capital took on part of a coaching role: explaining how investment institutions evaluate startups, accompanying Li Jianxiong in meetings. Li Zhengwei recalls that early interactions were dominated by Professor Zhan, with Li Jianxiong speaking very little. It was only after deeper engagement that they realized Li Jianxiong was actually leading the company—his silence stemmed from lack of fundraising experience.

"After meeting about five investment firms, things started to click," Li Zhengwei said. "He learns quickly and absorbs feedback rapidly, adjusting himself accordingly."

Fundraising soon became an intense, full-time effort. On his busiest day, Li Jianxiong met with five or six investor groups from 8 a.m. to 8 p.m.

Yet, meeting frequency didn’t equal progress. One firm promised disbursement next month—excitement ensued—but ultimately withdrew. Another firm that expressed interest also pulled out during final decision-making.

Li Jianxiong describes the fundraising process as "riding a pirate ship": today several investors show interest, tomorrow someone exits; the third day brings new investors, the fourth day someone else drops out.

The question investors ask most frequently is about commercialization. Li Jianxiong told China Entrepreneur that 99% of investors seriously inquire: Who is your target customer? How many orders do you plan to secure in the next few years?

He sees this as a disconnect between market expectations and the current stage of technological development.

"Our rationale for founding a company in embodied intelligence stems from the belief that commercialization is premature. The technological paradigm hasn’t yet reached a state suitable for large-scale deployment. Choosing commercialization now is essentially 'building a house with chopsticks'—drinking poison to quench thirst," Li Jianxiong explained. He aims to let commercialization emerge naturally through fundamental breakthroughs in the underlying technology paradigm.

Only 1% of investors strongly advise him: "Be yourself. Don’t talk about commercialization. Focus on doing your work well. Let time handle the rest—go with the flow."

One comment deeply impressed Li Jianxiong: "Embodied intelligence is an industry with an extremely long deployment cycle—perhaps 20 years before achieving large-scale domestic rollout. By then, you’ll be 47, exactly at the peak of societal contribution. So if you're investing in embodied intelligence, you should invest in you."

WuYuan Capital was Li Jianxiong’s third deep contact and the first institution to actually invest. During their initial conversation, the investor noticed Li Jianxiong held independent views on the technology roadmap—rather than parroting established industry narratives. This resonated with the investor’s preference for "classic startups"—smaller teams, limited resources, but tackling bold ventures that big tech or others dare not attempt.

Self-Evolving "Average Student"

Li Jianxiong defines his growth journey as that of an "average student":

At elementary school, his grades were average. After lucky admission to a key middle school, he entered a key high school as the last-ranked transfer student. He barely scraped into Xi’an Jiaotong University in the college entrance exam. Even in university, he never saw himself as among the top performers.

Because he didn’t believe he was superior, he willingly invested more time. During undergrad, studying mechanical engineering, he joined robotics competitions. His performance gradually rose to first place in his major, earning him a direct admission to Tsinghua University’s PhD program.

Initially, he contacted the School of Vehicle and Transportation at Tsinghua University, hoping to make vehicles or other machinery "intelligently move." Due to fixed admissions slots with his preferred supervisor, he was redirected to AIR, newly established under the leadership of Dr. Andrew Yao.

Shifting direction meant rebuilding his entire skillset almost from scratch. Upon receiving AIR’s acceptance, he hadn’t yet used Python or trained basic neural network frameworks. He taught himself by reading books, deriving formulas, and replicating code.

It wasn’t until 2023 that he published his first AI-focused academic paper. Dr. Yao advised him: don’t fear short-term lagging—trust your ability to run a marathon.

Photography: She Guisen

After entering embodied intelligence research, his original mechanical engineering skills proved invaluable. Early on, no one in the team truly understood mechanics—a simple camera tripod once stalled the entire project. Li Jianxiong stepped in directly—designing, fabricating, and integrating systems. In the lab, whenever a module failed to turn, he’d jump in personally to get the system running.

To WuYuan Capital’s investor, Li Jianxiong possesses a unique charisma forged by technical excellence, capable of uniting talented young minds. When asked why team members followed him into entrepreneurship, responses included: "He’s incredibly reliable—no procrastination at all." Projects led by Li Jianxiong were typically completed days ahead of schedule, not waiting until the last minute.

Yet as an entrepreneur, Li Jianxiong remains in transition—from "prodigy" to "mature founder." "Technical prowess doesn’t automatically imply understanding of organization and management," Li Zhengwei said.

Beyond fund allocation, Light Source Capital engaged Li Jianxiong frequently on team management, equity distribution, and commercialization strategy. Li Zhengwei’s insight: much in entrepreneurship is irreversible once decided—especially equity splits. "Once set, there’s almost no room for adjustment afterward."

Li Jianxiong once claimed he didn’t care about personal gain, initially proposing ~10% equity for each founder, believing this fair. But investors saw this as flawed: "Equal distribution weakens the CEO’s incentive and decision-making power."

"Equity touches human nature, responsibility, and motivation—different from scientific research logic," Li Zhengwei said. "Investors want to see a CEO fully committed, fully incentivized, and fully empowered."

Post-entrepreneurship, the volume of information he must process has surged. Formerly focused only on whether research was worthwhile, he now grapples with organizational structure, partnerships, and the company’s strategic trajectory—pondering how to advance while staying true to his original mission.

"I’m a bit like In-context learning—I’m undergoing my own on-device self-evolution," Li Jianxiong said. He hopes, through continuous trial and error, to broaden and deepen his thinking—to evolve from a technologist obsessed with breakthroughs into a true entrepreneur.

Source: China Entrepreneur Magazine