Grad-school DDPG that teaches drones to offload edge tasks

What it does This repository contains the Python implementation behind a Springer-published paper on wireless networks. It models a drone as a flying edge server that accepts partial task offloads from nearby ground devices, then uses a Deep Deterministic Policy Gradient agent to jointly optimize user scheduling, offloading ratios, and the UAV’s flight angle and speed. The goal is to minimize the maximum processing delay across all users while respecting energy and discrete-variable constraints.

The interesting bit Most edge-computing heuristics separate flight planning from task scheduling; this work fuses them into a single continuous-control problem. The DDPG agent simultaneously decides user scheduling, offloading ratios, and the drone’s flight angle and speed—an unusual marriage of actor-critic methods and wireless resource management.

Key highlights

• Direct implementation of the published Wireless Networks paper, citation and all.
• Jointly optimizes user scheduling, task offloading ratio, UAV flight angle, and flight speed as a single policy.
• Uses DDPG to handle the continuous action space and non-convex constraints that DQN struggles with.
• Ships as a reference implementation in Python for TensorFlow 1.x.

Caveats

• Explicitly unmaintained: the original graduate-student author no longer works in research, so updates are unlikely.
• Locked to TensorFlow 1.x, which is now well past end-of-life.
• Provided strictly as a reference implementation; do not expect polished packaging or active support.

Verdict Worth a look if you are researching UAV-assisted edge computing or need a concrete, cited example of applying DDPG to network resource allocation. Skip it if you want a maintained, modern framework—this is a frozen grad-school artifact, not an active product.