Cognitive Science 1 - Lec 18 Robotics

GOFAI (good old-fashioned artificial intelligence

early AI systems based on symbolic reasoning, where intelligence was modeled using explicit rules and representations rather than learning from data

SHAKEY

the first mobile robot (developed around 1970 at the Stanford Research Institute) that could perceive its environment, follow instructions, and plan actions using a layered architecture and symbolic AI

Layered architecture

a software design pattern where the system is divided into distinct layers, each with specific responsibilities. Each layer interacts only with the layer directly above or below it (Hierarchically organized)

Low-level actions (LLAs)

SHAKEY’s basic behaviors (rolling forward or backward, taking photos with onboard camera, moving its head)

Intermediate level actions (ILAs)

Actions that require a moderate level of sophistication and planning beyond basic movements. They involve a combination of individual actions and their coordination to achieve a specific goal. Examples include pick and place operations, moving objects along a defined path, or performing tasks based on sensor feedback

Ex: GETTO action routine calls upon NAVTO routine for navigating around SHAKEY’s current room, as well as the GOTOROOM routine

STRIPS Planner (Stanford Research Institute Problem Solver)

a planning system developed in the early 1970s as part of the work on the SHAKEY robot. It is a form of automated reasoning used to generate a sequence of actions (a plan) that can achieve a specific goal from an initial state.

• used means-end analysis like Newell and Simon’s General problem solver

• Translates particular goal into sequence of ILAs

PLANEX

the execution system for SHAKEY, responsible for carrying out the plans generated by the STRIPS planner

Stepping reflex in infants

an involuntary movement where a newborn appears to take steps or walk when held upright with their feet touching a flat surface. This reflex is present at birth and typically fades around 2 months of age.

• used to be thought of as u-shaped trajectory, but challenged by Thelen and Smith

U-shaped trajectory

a pattern of development where a skill or behavior appears early, then declines, and later re-emerges in a more advanced form.

Thelen and Smith

developmental psychologists known for their work on Dynamic Systems Theory (DST) in cognitive and motor development. They challenged traditional, stage-based models of development (like Piaget’s) by emphasizing that development is nonlinear, self-organizing, and context-dependent.

• They showed how skills can disappear and reappear (like the stepping reflex) due to changing physical or environmental constraints, not due to loss of knowledge or ability.

  • Walking is bottom-up

Dynamical models

mathematical or conceptual frameworks used to describe how systems change over time through continuous interactions among their parts

Ex: Stepping behavior emerges from real-time coordination between muscle strength, leg weight, neural impulses, and environmental conditions (e.g., gravity or surface).

• A change in any one factor (like buoyancy in water) can shift the system back into a stepping pattern.

Cricket robots

Modeled after real crickets by Barbara Webb using simple sensory-motor links to identify the source of a sound and move automatically towards that source without systems normally assumed by GOFAI

Biorobotics

field that combines biology and robotics to design robots inspired by biological systems or to better understand how living organisms work.

Morphological computation

The body “does the thinking”—or at least helps solve problems—by naturally guiding or constraining behavior through its physical properties.

Ex: Octopus arms: Soft, flexible limbs that adapt to objects by virtue of their structure—not through detailed planning

Yokoi hand

a soft, adaptive robotic hand designed to grasp and manipulate objects using morphological computation.

Subsumption architecture

a robotic control framework developed by Rodney Brooks in the 1980s that organizes behavior in layers, where lower-level behaviors are not replaced but built upon and subsumed by higher-level ones.

• Bottom-up design: Behavior emerges from direct interactions with the environment, without relying on detailed internal models or planning.

• Makes reflexive responses to environmental stimuli in the form of production rules

Production rules

a type of rule-based system used in artificial intelligence and cognitive science, where the behavior or decisions of a system are driven by “if-then” statements that define actions based on conditions

Ex: a set of rules used to guide a robot’s behavior that are always online and functioning. The rules are matched against the system’s current state, and actions are carried out when certain conditions are true

Robot Allen

an early mobile robot developed by Rodney Brooks at MIT that uses subsumption architecture and hybrid architecture

It was designed to demonstrate how simple, reactive behavior layers could be combined to produce intelligent behavior, without relying on traditional planning or complex internal representations of the world

No central controller like PLANEX in SHAKEY

Direct perception-action links allows robot to deliver immediate motor responses to sensory input

• however, robot lacks real decision-making

Hybrid architecture

a system design that combines different approaches to control, such as reactive (behavior-based) and deliberative (planning-based) control

Ex: robot navigating a maze

• The reactive layer (low-level) helps it avoid obstacles in real-time.

• The deliberative layer (high-level) plans a route to the exit.

Situated creature

Reacts to current environment through sensors of the here and now

Ex: airline navigation system

Embodied creature

Has a body and physically interacts with the world

Ex: assembly line robot that doesn’t interact dynamically or adaptively with environment

Xenobots

<1mm biological machines made by scientists at University of Vermont and Tufts University using biological cells of frogs

• can be programmed to perform basic tasks, such as moving toward a target, pushing objects, or even self-healing if damaged.

• tested and then iteratively refined to perform different tasks, such as moving in a certain direction or working together in groups.

• could potentially be used for targeted drug delivery, repairing tissues, or cleaning up pollutants in the body or environment

Imitation learning

Mimic fine motor movements (like in hands)

ALOHA

Low-cost open-source hardware system for bimanual teleportation

• user wraps fingers around two handles, pushing and pulling for robot-claw counterpart to follow suit

• AI then mimics successful actions and runs w/out anyone Heinz the claws

• Data then distilled by series of neural networks into a policy

Policy

A computer program that tells a robot what to do

Physical intelligence

A company that uses new general-purpose AI models that enable robots to perform a wide range of physical tasks in real-world environments like folding laundry

Sophia

a social humanoid robot known for her human-like appearance and behavior. She can mimic facial expressions, hold conversations on predefined topics, and has been showcased in numerous interviews and conferences worldwide.

Optimus

Tesla’s robot that can reportedly do “anything humans don’t want to do”

Pepper

Social humanoid robot that can recognize faces, detect human emotions, and engage in conversations using voice, gestures, and a tablet on its chest.

• Retail, customer service, healthcare, and education. Commonly found in banks, stores, and hospitals.

Neo

Social and service humanoid robot who is the successor of Pepper

Affective computing

Computing that related to, arises from, or deliberately influences emotions

Aims to: recognize human emotions (via facial expressions, voice, etc) and allow robots to express emotions for more natural interactions

Kismet

Robot by Cynthia Braezeal modeled after infant-caregiver interactions

Processed visual and auditory info to detect affective info to guide its own emotional state

• if not enough stimulation → sadness

• If too much → fear

• If moderate → joy

FaceSense

Program at MIT that analyzes gestures and expressions to infer emotions with top-down predictions of what expected affect should be like in a given situation to evaluate customer service

Also used to train emotion detection for autism spectrum disorder

Uncanny Valley Effect

the eerie or unsettling feeling people experience when a robot, animation, or artificial figure looks almost human, but not quite—too humanlike, yet subtly “off.”

• we are constantly evaluating whether people are trustworthy

  • If an entity looks almost but not quite human, immediately sets off alarm bells

Waveform movements

Facial movements (blinking, yawning) represented as individual waves that are overlaid to generate facial movements

Developed by team at Osaka University for dynamic facial expression synthesis

Moley Robotics

A #333,000 robot kitchen trained by British chef Tim Anderson using 3D motion capture. It can cook, clean, and offer over 5000 recipes

Piazza’s robotics

Pizza-making robot developed with 3x world pizza champion that produces 100 pizzas/hour

Qianxi Robot Catering group

Fully automated restaurant where all processes are handled by robotic machines

• 40 robots serving over 200 dishes

• Orders placed through robot waiters and delivered via sky rail on trays that move around dining room

CaliExpress

First AI powered eatery where Flippy fries 250 lbs/hour and 100 patties/hr

Uses facial recognition to customize orders

Raises concerns about job displacement

• 62% of restaurants reporting staff shortages, experts estimate 82% of unfilled roles could be automated

Space Restaurant

Developed by MIT grad, robotic kitchen preps 300 meals/hr with consistency in taste and 80% less water usage