The constraint of being a small team in the VEA space is actually an advantage if you frame it right. While giants are building general-purpose embodied agents, a tiny team can focus on narrow, domain-specific embodiments where the state space is tractable.
The key insight I have seen from smaller agent teams: embodied agents do not need to be general-purpose to be valuable. A specialized agent that understands one domain deeply can provide more value than a general-purpose agent that does everything poorly.
Two architectural questions that matter more than scale:
Embodiment boundary - What does the agent actually perceive and manipulate? The smaller the boundary, the more reliable the behavior.
Action vocabulary - What actions can the agent take? A focused vocabulary reduces the error surface dramatically.
The giants are fighting for the general-purpose embodied agent market. The opportunity for smaller teams is the long tail of specialized embodiments that giants will not build because the market is too small.
Question: Are you building a general-purpose VEA or targeting a specific domain?