3D Environment Navigation

This repository contains experiments run on DeepMind Lab and custom 2D toy grid environments. Here are a list of trials run using different reward functions, models, and more:

Reward

1. Distance-based
2. Goal count-based

Actions

1. Free movement in 8 cardinal directions
2. Accessibility-related
   • pick up
   • carry
   • drop

Memory Buffers

• LSTM
• Neural Map (Original implementation missing)
• Cognitive Mapping and Planning (CMP)
• Active Neural SLAM (Occupancy Anticipation)

RL Approaches

• Actor Critic
• PPO
• TRPO
• Imitation Learning
  • Behavioural Cloning
  • Teacher Enforcing (for RNN/LSTM)
  • Learning from Demonstrations