Roundtable Discussion on Embodied Intelligence at Meituan Robot Research Institute 2024 Academic Annual Meeting

In 2024, embodied intelligence became a highly anticipated field, representing the future of human‑machine relationships. To explore this topic, Meituan Robot Research Institute held a round‑table forum at its academic annual meeting, inviting several experts.

Panelists : Tan Zhangxi – Founder & CEO of Ruisi Xinkechuang and Director of RIOS Turing Lab; Wang He – Assistant Professor at Peking University, Founder & CTO of Galaxy General, Zhiyuan scholar; Wu Yi – Assistant Professor at Tsinghua University Institute of Interdisciplinary Information; Zhao Xing – Assistant Professor at Tsinghua University and Chief Scientist of Xinghaitu; Ding Wenbo – Associate Professor at Tsinghua Shenzhen International Graduate School (moderator).

Ding Wenbo (moderator) : Opened the session, asking each guest to briefly introduce their background and recent work related to embodied intelligence.

Tan Zhangxi : Spent the past five years on chip development (RISC‑V open‑source compute chips) and a sensor‑compute startup sold to Aurora. Emphasized the importance of hardware foundations for embodied AI.

Wang He : Described his long‑term research on embodied intelligence from Stanford to Galaxy General, positioning the company as a leader in large‑model robotics in China. Highlighted the need for both hardware (body) and software (intelligence) and argued that the current bottleneck shifts over time.

Wu Yi : Focuses on multi‑agent reinforcement learning, agent reasoning, and human‑robot collaboration. Discussed the challenge of giving robots spatial understanding and long‑term memory to perform everyday tasks like locating objects.

Zhao Xing : Shared experience from Waymo and current work at Xinghaitu. Stated that the “body” of a robot is already mature; the main challenge lies in intelligent perception, reasoning, planning, and control, especially leveraging large‑scale pre‑training.

Ding Wenbo : Summarized the panel’s backgrounds, noting a mix of chip, hardware, and learning expertise, and introduced the first thematic question: which is more critical for embodied intelligence, the body or the intelligence?

Wang He : Argued that both are inseparable; the bottleneck now is intelligent algorithms rather than hardware.

Zhao Xing : Emphasized that commercial viability depends on closing the loop between data, models, and market; safety and reliability are also essential.

Tan Zhangxi : Highlighted supply‑chain fragmentation and the need for a mass‑market application (akin to iTunes for MP3) to drive large‑scale adoption.

Wu Yi : Stressed that embodied intelligence should ultimately serve diverse user groups, including the elderly, and that long‑tail, emotionally supportive services are important.

Discussion on Market Opportunities : The panel identified several promising domains – hazardous or inaccessible environments (high‑altitude, underground, border work), household assistance, and large‑scale retail/warehouse pick‑and‑place tasks. They agreed that a 5‑year horizon could see widespread deployment in dangerous jobs, while a 10‑year horizon may open the home market.

Human‑Robot Co‑existence : The speakers debated safety, compliance, and the level of physical interaction humans would tolerate. Examples included data‑driven hand‑over tasks, force‑controlled manipulation, and the need for small, unobtrusive robots before scaling to full‑size humanoids.

Future Milestones : Wang He projected 10,000 humanoid robots operating autonomously within five years, likening the trajectory to that of electric vehicles. Zhao Xing highlighted the importance of affordable, mass‑produced products as a sign of a mature industry.

Closing Remarks : Ding Wenbo reflected on the “embodied intelligence era” as a time for self‑breakthrough and evolution, acknowledging both the excitement and the ethical concerns of creating self‑improving machines.

The round‑table concluded with gratitude to all participants.