This project won 1st place worldwide in robotics in Regeneron ISEF 2023 (International Science and Engineering Fair), won the SAE special award in Regeneron ISEF 2023, and won 1st place in MSEF (Massachusetts Statewide Science and Engineering Fair)

This project of mine is a precision rocket landing algorithm that can give instructions to the engines and the thrusters of the rocket every 0.1 seconds during atmospheric entry to control the descent of the rocket and land it within two meters of a specified target landing location on interplanetary missions such as Mars missions.

The precision rocket landing algorithm first uses a convolutional neural network (CNN) algorithm that can determine a safe landing spot for the rocket (surface without rocks, hills etc.) given an image of the terrain underneath, available to the rocket. The coordinates of this safe landing spot is given to a model predictive control algorithm. The model predictive control algorithm generates a tree of different actuation sequences (scenarios) that the rocket can choose to implement, and uses a nonlinear physics model of the rocket to predict the future states of the rocket given each actuation scenario. A cost function then evaluates each scenario based on how close it is to safe landing, and the optimal scenario of actuations is chosen. The first step of the optimal actuation sequence is fed into the engines and thrusters of the rocket every 0.1 seconds in the form of instructions.

After testing the precision rocket landing algorithm under various conditions in Mars-like landing situations, it was determined that the algorithm is able to achieve a 100% successful landing rate within 250 meters of required trajectory correction. In more extreme conditions, it was able to achieve 100% successful landing rate within 150 meters of trajectory correction. The final version of the algorithm was 1000+ lines of code, but I wrote more than 15000 lines of code in the totality of the project.