Large-Scale DRL Exploration with Tunable Adaptive Exploration-Exploitation

Overview

Deep reinforcement learning-based reactive planner for large-scale Lidar-based autonomous robot exploration with tunable exploration-exploitation control.

• Optimized distributed Soft Actor-Critic (SAC) training through comprehensive hyperparameter tuning with experience replay buffer optimization and adaptive learning rate scheduling

• Engineered Alpha Conditioning system with continuous parameter control for precise exploration-exploitation behavior modulation in robot navigation policies

Technical Stack

Tech Stack: PyTorch, Ray, Soft Actor-Critic (SAC), Graph Neural Networks