15/06/2026
Autonomous Rocket Lander - Summer Internship Project 2025
What does it take to teach a rocket how to land itself?
Using Deep Reinforcement Learning (PPO), we trained an AI agent to perform controlled rocket landings by learning from trial and error in a simulated environment.
Key Highlights
Continuous Control
The agent learns to manage throttle, side thrust, and nozzle angle in real time.
Rich State Information
Position, velocity, orientation, angular velocity, and landing-leg contacts are used to make decisions.
Reward Design
Rewards encourage smooth, precise, and fuel-efficient landings while penalizing crashes and unstable behavior.
Multiple Landing Scenarios
The rocket was trained on both stationary and moving barges to improve robustness and adaptability.
Results
From random actions to reliable soft landings, the agent gradually learned stable descent strategies and achieved consistent performance after training.
Deep Reinforcement Learning in action - transforming exploration into intelligent control.
Team Members:
Raj Patil, Tvisha Mehta, Ishan Agrawal
Mentors:
Akash Tiwari, Atharv Kulkarni, Aarush Sinha, Anuj Sidam
