Skip to content

Warren 2006

The Dynamics of Perception and Action

Tags:

Related:

Key Definitions

Behavioral Dynamics:

Dynamical Hypothesis: The morphology of human and animal behavior can be formalized in terms of low dimensional dynamical systems

State space:

Dynamical System: A system of first order differential equations (ODEs) in which the rate of change of each variable is a function of the current state of the system. In this context they describe movement in state space, and the actual behavior corresponds to solutions to the system of ODEs for a set of initial conditions.

Attractor: Location in state space towards which trajectories converge from different sets of initial conditions.

Repeller: Location in state space away from which trajectories diverge from different sets of initial conditions.

Bifurcation: Sudden change in the number or stability of attractors or repellers in state space coming from a continuous change in system parameters.

Dimensionality: The minimum number of state variables sufficient to predict the future state of the system.

Linearity: Whether a dynamical system contains nonlinear terms or not.

Stability: The stability of a point in state space is the slope of the curve at that location

Control Parameter: The parameter whose change creates a bifurcation

Key Takeaways

  1. The primary question is how behavior is organized, which is roughly how do humans tightly organize their behavior to the environment while achieving a specific goal. This requires perception in order to use information about the world and body to choose the correct action given the goal and environment, and the coordination of the body into action.
    1. Commonly it is thought that the organization of behavior implies centralized control over behavior. Different theories have attributed different qualities to this supposed control:
      1. Neuroreductionist view: organization of behavior is due to prior organization in the structure of the nervous system.
      2. Cognitivist view: organization of behavior comes from the structure of internal representations of the world and body
      3. Behaviorist view: organization of behavior is just response to contingencies in the environment
    2. Generally, these simply beg the question of the organization of behavior to the organization of some preexisting internal or external structure.
    3. "Behavior is regular without being regulated" - Gibson 1979
    4. Another option, first proposed by Gibson, is that behavior emerges from the constraints of "the structure and physics of the environment, the biomechanics of the body, perceptual information about the state of the agent-environment system, and the demands of the task."
    5. The resulting solution must be able to explain the generally stable and reproducible quality of human movement, as well as its flexibility and adaptability.
  2. Two levels of analysis for adaptive behavior:
    1. Perception and Action: Agent and environment are mutually coupled dynamical systems. This coupling is mechanical as well as informational. The agent-environment interactions cause emergent behavior with its own dynamics, which is subject of the second level.
    2. Behavioral Dynamics: The time evolution of behavior can be represented as a vector field, where stable behavioral solutions correspond to attractors and transitions between behaviors are bifurcations. Stabilities are codetermined by the confluence of task constraints and perceptual-motor control laws, rather than existing a priori in either the agent or the environment. Control exists in agent-environment system, as it is determined by the constraints of both dynamical systems. Behavior patterns emerge from bootstrapping process (learning, development, evolution) where agent-environment interactions give rise to behavioral dynamics, and stabilities in the behavioral dynamics affect the behavior of the agent. This can be thought of as the agent exploring the vector field of a task, looking for stabilities while also acting to create them. From the agent's point of view, this looks like exploiting physical and informational constraints to stabilize the intended behavior.
  3. The scale of analysis here is the scale of the observed regularity, not at micro levels. "A theoretical account of behavior must incorporate goals, information, physics, and properties of the world that are not directly reducible to a neural level."
  4. Based on various previous theories based on both representational and non-representational approaches, four themes of a promising approach to the dynamics of perception and action are noted:
    1. Embodiment and embeddedness: The agent has a physical body and lives in a physical environment, both of which constrain behavior (contrast to Fodor style: agent can only think in representational model and affect representational model)
    2. Information based control: Behavior is guided by the information the agent has about itself and the environment
    3. Task specificity: Control is task specific, based more on special case solutions than a successful generally applicable model
    4. Emergent self organized behavior: Behavior emerges from interactions between agent and environment; is constrained by the physical world, information, and task; and self organizes through feedback towards stable action patterns
  5. The dynamical hypothesis says that human and animal behavior can be formalized in terms of low dimensional dynamical systems, with stable modes of behavior being attractors and changes in behavior being bifurcations. The dynamical system is a set of first order differential equations describing motion, the solutions to which are the exhibited behavior. This constricts the morphology of behavior to a limited set of low dimensional fixed points, limit cycles, strange attractors, and bifurcations, making it easier to formulate specific theoretical claims.
    1. The laws of physics make up a low dimensional dynamical system describing changes in the environment, and control laws make up a low dimensional dynamical system of the action of an agent, dependent on neuromuscular and biomechanical degrees of freedom as well as information available to the agent.
    2. Agent and environment are coupled in two ways: The agents actions form an effector function that changes the environment, and changes in the environment form an information function that changes the information received by the agent. These make up the perception-action cycle.
    3. Adaptive, goal oriented behavior arises from the perception action cycle, and the time evolution of this behavior is then characterized formally through behavioral dynamics.
    4. In this setup, the nervous system changes the mapping from information variables to control laws in order to produce stable behavior. But the behavior is also influenced by the environment, and thus the nervous system does not have total control.

Limitations