Behavior-based Robotics

Behavior-based Robotics

Overview of Robotics Approaches

• Two main approaches to robotic AI systems are:
  • Model-based Robotics: Intelligence with internal representations.
  • Example: ACT-R (Adaptive Control of Thought – Rational)
  • Behavior-based Robotics: Intelligence without internal representations.
  • Example: BigDog, Roomba

Model-based Robotics/AI Systems

• Key Features:
  • Relies on symbolic processing and representation of knowledge.
  • ACT-R was developed by John Anderson at Carnegie Mellon University (CMU) in the early 1990s.
  • Based on the tradition of the General Problem Solver (GPS).
  • Focuses on understanding how humans organize knowledge and exhibit intelligent behavior.
  • ACT-R contains specialized modules responsible for perception, memory, and action, which interact through buffers.
  • Each module is associated with specific brain regions, supporting specialized functions in cognition.

ACT-R Cognitive Architecture

• Modules in ACT-R:
  • Perceptual-motor Module: Deals with sensory inputs and motor actions.
  • Goal Module: Manages the current objectives.
  • Declarative Memory Module: Stores information and past experiences.
• Working Mechanism:
  • Chunks (units of knowledge) are placed in buffers for access by the production system, which applies rules to these chunks based on the current information pattern.
  • Learning occurs through the tuning of subsymbolic processes guiding rule selection.

History and Evolution of ACT-R

• Initial development focused on the integration of various modules to produce coherent cognition.
• Continuous refinement has allowed ACT-R to evolve closer to performing a vast range of human cognitive tasks.

Behavior-based Robotics/AI Systems

• Key Features:
  • Operate based on real-time sensory inputs with no need for internal representations.
  • Reacts to the environment directly rather than manipulating internal symbols.
• Example: The Cricket Robot that responds to sound intensity without an internal map of its world.
• Developed by Rodney Brooks of MIT, known for introducing the concept of intelligent behavior emerging from environmental interactions rather than complex cognitive mapping.

Subsumption Architecture

• Proposed by Brooks, consists of multiple layers of behavior that operate simultaneously.
• Lower layers handle basic reflexive actions, while higher layers build upon these to exhibit more complex behaviors (e.g., exploration).
• Demonstrated through robots like ALLEN:
  • Layer Functions:
  • Layer 1: Avoid collisions
  • Layer 2: Wander freely without hitting obstacles
  • Layer 3: Search for interesting environments

Comparison of Robotics Approaches

• Model-based Robotics (e.g., ACT-R)
  • Symbolic processing, slower response, high-level intelligence.
• Behavior-based Robotics (e.g., Roomba, BigDog)
  • Reflexive processing, real-time response, low-level intelligence.

Discussion Questions

1. Is it essential for the robot ALLEN (subsumption architecture) to inhabit a real world instead of a virtual one, and why? How does this apply to SHAKEY?
2. How would you characterize the mammalian brain: as a subsumption architecture, a modular architecture, or a hybrid model?

Important Notes for Final Exam

• The final exam is on April 18, from 2:20 PM to 3:40 PM.
• All exams are online and browser-locked, requiring physical attendance in class for participation.
• Ensure to bring your BuckID and familiarize yourself with the exam regulations regarding rescheduling and late attendance.