Document Type
Thesis
Date of Award
5-31-2026
Degree Name
Master of Science in Electrical Engineering - (M.S.)
Department
Electrical and Computer Engineering
First Advisor
Arnob Ghosh
Second Advisor
Adeel Akhtar
Third Advisor
MengChu Zhou
Abstract
Research in autonomous exploration has created many effective algorithms that have been tested and proven to work in many different virtual and physical environments. Many optimizations have also been developed to reduce computational effort and increase exploration speed.
However, despite optimizations, these algorithms can still require considerable computational effort and time to explore even small environments. To obtain further improvements in computation and exploration speed, a reinforcement learning agent using actor-critic style proximal policy optimization (PPO) is trained to explore various environments efficiently, then compared to an algorithm using contemporary exploration methods.
Testing is performed in virtual environments with ideal sensor data, then in physical environments by reconstructing virtual environment layouts in the real-world. Virtual results show the agent's computation and exploration speed is much faster than the baseline algorithm while still being move efficient, but physical results reveal the agent's weakness to noisy environments caused by external factors like LiDAR noise, positional drift, rotational drift, map distortion, etc.. The agent could not be trained on these factors in time for this paper.
Recommended Citation
Knight, Brandon, "Autonomous exploration of an environment with static obstacles using a ppo agent" (2026). Theses. 3552.
https://digitalcommons.njit.edu/theses/3552
