Learning Symbol Groundings via Simulated Robotics

1
Learning Symbol Groundings via Simulated Robotics

• Ari Heljakka
• GenMind Ltdjoint with
• Novamente LLC
• MAY 2006

recommend
2
Overview

• The Problem Grounding of symbols
• Nouns (simple objects)
• Prepositions (relations, complex objects)
• Solution
• Virtual Embodiment for integrated AI
• Novamente

3
The Plan

• Ground (Virtual Reality)
• AGI-Sim
• Concept mining
• MOSES (Estimation of Distribution)
• Clustering
• Abstraction / Inference
• PLN (Probabilistic Logic Networks)

4
Grounding Nouns

• AGI-Sim perception data
• Visual
• Words
• Word-object association via
• Object recognition (Concept mining)
• Frequent pattern miner
• Elementary PLN inference

5
AGI-SimAn Open Source Simulation Environment for
AGI

• AI systems can sense and act in real-time via
  embodiment in a 3D virtual world
• Uses CrystalSpace (open-source game engine) for
  visualization
• Provides AI systems with inputs from various
  built-in sensory modalities
• Integration with natural language interface for
  fluid, situated communication
• Suitable for teaching/learning by methods from
  developmental psychology
• Usable by any AI system via a simple
  sockets-based protocol

6
AGI-Sim Video
7
Probabilistic Logic In Novamente

• High level
• Relations between high-level nodes
• Tasks planning
• Low-level
• Control of motoric schemata
• Use knowledge from other agents (eg .clustering)
• Predicate schematization

8
PLN Word-object Association Inference 1/2
9
PLN Word-object Association Inference 2/2
10
Grounding Prepositions

• AGI-Sim
• SMEPH
• PLN Engine
• Backward-chaining
• Forward computation

11
(No Transcript)
12
Future Development

• Engine
• Mining proof tree statistics
• PLN-MOSES integration
• Attention allocation (priority heuristics)
• Background inference
• Grounding prepositions of increasing complexity
• Climbing Piaget phases

13
(No Transcript)