Skip to content

Link to Embodied Intelligence

This chapter is the bridge where this chapter meets the Human-Like Intelligence chapter.

The Human-Like Intelligence chapter is about "how to build a mind on the digital side" — world models, predictive coding, causal reasoning, meta-learning. This chapter now reveals a deep mapping: the neural manifolds studied by BCI research are essentially the latent space of a biological policy; motor cortex as a dynamical system is essentially an online-learning RL policy network.

Put differently, BCI lets us for the first time "read out" a working world model and policy network from real biological neural systems. This provides biological empirical validation for purely algorithmic theories like world models and JEPA.

Why this chapter sits at 10 rather than earlier. Only once the neural manifolds of Chapter 02, the decoders of Chapters 04–05, and the I2A paradigm of Chapter 06 are firmly in place can the mapping "motor cortex = online RL policy" be more than a slogan. This chapter teaches no new algorithm; instead it draws lines between concepts you have already learned: read Churchland-Shenoy's rotational dynamics as policy smoothness in RL, read neural manifolds as JEPA's latent space, and read BCI's closed-loop training as brain-in-the-loop validation. These connections have direct value for anyone working on world models, embodied intelligence, or RL.

Recommended reading order. Motor Cortex as a Dynamical System and Neural Manifolds and RL Policy are the two core sections — read them first. Then BCI as World Model Validation introduces the brain-in-the-loop experimental paradigm (this is the most direct response to the Human-Like Intelligence chapter). Finally Brain-Body-Environment Closed Loop lands the abstract mapping in concrete systems via the Walk Again Project and 2025 progress on bidirectional exoskeletons.

Chapter contents:

评论 #