Physix Frontier · News Briefing Card (enbrief · Nov 14, 2026)

ZJU's Peng Sida shifts from 3D reconstruction to embodied AI navigation

KEY FACTS

  • Peng Sida stayed on at Zhejiang University after his 2023 PhD to become a researcher at the School of Software Technology.
  • His team publicly released two robot navigation works, SuperNav and ScaleNav, in October.
  • SuperNav handles decision-making and planning, while ScaleNav handles low-level locomotion control.
  • Street Gaussians has been applied to Li Auto's autonomous driving simulation system.
  • Peng Sida argues that in embodied intelligence the action modality matters more than the visual modality.

PHYSIX OBSERVATION

Peng Sida's choice of route is worth noting: rather than chasing the end-to-end large-model hype, he pairs a general-purpose VLM with an Agent that bolts on cerebellum-style tools, sidestepping the communication-latency bottleneck of local large-model deployment. This is a reminder for embodied AI startups: stacking parameters is less useful than cleanly separating perception, memory and action, and engineering feasibility may determine how fast you ship more than model scale does.

Source: enbrief original report ↗