Lucas Rouchy

Reinforced Learning Based Controls and Planning

09/2022 - Present

OSU Global Formula Racing Team - FSAE

This project aims to implement a deep reinforcement learning policy using camera and LIDAR data to discover our inputs and produce safe outputs for steering and throttle actuators.

• Implemented a distributed training model using the Ray package.
• Integrated the distributed model into our PyTorch implementation.
• Documented each progression for future teammates and to ensure maintainability.
• Constructed a new two-layer reward policy with colleagues, improving simulated track performance by over 300%.
• Simplified training environment while keeping randomness through use of random changes to the state of the agent.

Ecosystem Simulation

04/2023-05/2023

Developed a simulation of an ecosystem with a focus on the predator-prey relationships. The ecosystem is affected by precipitation and temperature.

Construction of the simulation was done in C plus plus with a focus on concurrency.

Business Reviews API

03/2023 - present

Developed a RESTful API for a business review website. The API is built with Node.js and Express.js.

Implemented a MongoDB database to store user and business information. Handling of datapoints is done using Mongoose.

The whole application is dockerized allowing for anyone to run the server with one command regardless of their environment.

Terminal Simulator and Memory Allocator

01/2022 - 02/2022

Developed an efficient command-line simulator through a test-driven development process. Replicated standard commands such as cd, rm, mkdir, nano, touch, cp, mv, and more.

Introduced commands: myalloc and free to allocate and free memory space. Development was done in C.

Multiprocess Generator

02/2021 - 02/2022

Program started as a population generator for the US with an Object-Oriented approach. Generator uses US census data over 10 years to feed the program.

Designed a geolocation content generator to create related pieces of content specific to user’s input. Content generator required implementing an interactive GUI.

These programs ran in parallel using multiprocessing techniques. Development was done in Python.