How divaksh06/fast-trained-latent-adaptive-quadrotor-control works
This repository is an early single-platform quadrotor hover-control RL baseline. Training creates or resumes a TD3 policy in a custom PyBullet hover environment; separate simulation and evaluation scripts run saved policies and report performance. The README also describes future latent adaptation and uncertainty-gated safety, but those modules are not present in the tree, so they are not modeled as implemented components. Arrows reflect only calls and data flow visible in the excerpts or README.
Main entry points: simulate_hover.py Hover diagnostics · paper_hover_env.py Paper hover environment · train_td3_paper.py TD3 training · eval_td3_paper.py Interactive evaluation
Components 8
Policy learning
- TD3 training
train/train_td3_paper.py - TD3 policy
train/train_td3_paper.py
Hover simulation
- Paper hover environment
envs/paper_hover_env.py - PyBullet drone simulator
Policy assessment
- Hover diagnostics
train/simulate_hover.py - Simulation reports
train/simulate_hover.py - Interactive evaluation
train/eval_td3_paper.py
Other components
- Researcher
Connections 11
- Researcher → TD3 training: starts training
README.md - TD3 training → Paper hover environment: creates environment
train/train_td3_paper.py - Paper hover environment → PyBullet drone simulator: steps physics
envs/paper_hover_env.py - TD3 training → TD3 policy: trains and saves
train/train_td3_paper.py - Researcher → Hover diagnostics: runs diagnostics
README.md - Hover diagnostics → TD3 policy: loads checkpoint
train/simulate_hover.py - Hover diagnostics → Paper hover environment: runs episodes
train/simulate_hover.py - Hover diagnostics → Simulation reports: generates
train/simulate_hover.py - Researcher → Interactive evaluation: runs evaluation
train/eval_td3_paper.py - Interactive evaluation → TD3 policy: loads checkpoint
train/eval_td3_paper.py - Interactive evaluation → Paper hover environment: runs episodes
train/eval_td3_paper.py