Vision to Auditory Substitution for an Artificial Agent
| dc.contributor.advisor | Menore Tekeba (PhD) | |
| dc.contributor.author | Semira Mohammed | |
| dc.date.accessioned | 2025-04-25T08:27:27Z | |
| dc.date.available | 2025-04-25T08:27:27Z | |
| dc.date.issued | 2025-01 | |
| dc.description.abstract | Sensory substitution technology converts raw visual input into auditory soundscapes, allowing individuals to “see” with sound. However, mastering this skill requires significant cognitive adaptation, extensive training, and practical application in realistic, everyday scenarios. Experiments with humans have shown the potential for auditory substitution of vision, but these efforts are limited by high costs, ethical concerns, and the risk of unintended side effects, such as impaired auditory skills. To address these challenges, this study develops a Vision-to-Auditory Sensory Substitution system for artificial agents. By simulating sensory substitution in a controlled reinforcement learning (RL) framework, this approach eliminates the need for human experimentation while retaining the ability to explore learning dynamics and decision-making behaviors. Using the Proximal Policy Optimization (PPO) algorithm, agents were trained in two OpenAI Gym environments—CarRacing-v2 and LunarLander-v2—to compare the performance of vision-based and auditory-based agents. The results demonstrate that auditory agents, despite inherent challenges in interpreting sound-encoded visual inputs, achieved means rewards of 427.91 in CarRacing-v2 environments and 259.85 in the LunarLander-v2 environment over 100 episodes. These findings highlight the potential of sensory substitution systems in enabling artificial agents to act effectively using auditory cues. This research contributes to advancing assistive technologies while addressing the limitations and risks of human-based sensory substitution experiments. | |
| dc.identifier.uri | https://etd.aau.edu.et/handle/123456789/5379 | |
| dc.language.iso | en_US | |
| dc.publisher | Addis Ababa University | |
| dc.subject | Machine Learning | |
| dc.subject | Proximal Policy Optimization | |
| dc.subject | Reinforcement learning | |
| dc.subject | Gym environments | |
| dc.subject | Vision-to-auditory Sensory substitution | |
| dc.title | Vision to Auditory Substitution for an Artificial Agent | |
| dc.type | Thesis |