NEW: Drone Learns To Complete Search & Rescue Missions
UAV Search & Rescue Simulation » Devlog
Environment
For training the agent, I used curriculum learning, which allows me to increase the difficulty gradually over training steps. For the curriculum, I decided to create three maps of differing sizes. A small, medium and large environment. The reason is that I want the agent to be able to handle travelling to farther places.
In previous training, I only trained on one map, which is the small one. This made the agent generalise the environment and become too uncomfortable, which makes the agent only know how to do the tasks in this one environment. I want it so the agent can scale well to many different environments no matter the size.

UAV Search & Rescue Simulation
Training Search & Rescue Drones with Reinforcement Learning.
| Status | In development |
| Author | Shivani |
| Genre | Simulation |
More posts
- NEW: UAV Search & Rescue Simulation ProgressAug 18, 2025
- Devlog #8 - Evaluating Trained ModelsJul 21, 2025
- Devlog #7 - Training My Drone: Curriculum Learning for Search TasksJul 15, 2025
- Devlog #6 - Training My Drone: Custom PPO ConfigurationJul 15, 2025
- Devlog #5 - Training My Drone: Battery AwarenessJul 08, 2025
- Devlog #4 - Training My Drone: Target Collection BasicsJun 26, 2025
- Devlog #3 - Training My Drone: ML-Agents SetupJun 26, 2025

Leave a comment
Log in with itch.io to leave a comment.