Session: 08-24-01: Human-Machine Interaction: Design, Dynamics, and Control

Paper Number: 165666

Robo-Brush: Autonomous Toothbrushing With Deep Reinforcement Learning and Learning From Demonstration 

For individuals with disabilities or limited mobility, performing daily personal tasks including eating, drinking, and performing hygiene tasks can be challenging and often require caregiver assistance. Automating these tasks using robotics can enhance independence and improve quality of life. However, achieving effective and adaptive robotic assistance poses significant challenges. Traditional control methods require extensive manual programming, while Deep Reinforcement Learning (DRL) offers promising solutions but demands high computational resources and long training times. To tackle these limitations, we introduce a hybrid learning approach that integrates Learning from Demonstration (LfD) with DRL, leveraging human guidance to accelerate learning while maintaining adaptability and autonomy. 

As a case study, we focus on an autonomous tooth hygiene task (i.e., toothbrushing) using a robotic arm that adapts to different dental structures and perform precise, effective toothbrush movement. We developed a physics-based simulation in the MuJoCo environment, modeling a Franka Emika Panda robotic arm, an electric toothbrush, and a human mouth with consisting 28 individual teeth. The robot's motion is obtained by an end-effector Cartesian control strategy, providing smooth and controlled brushing along predefined trajectories. The system was trained once only by DRL and another time by combination of LfD and DRL, where a dataset was collected by manually controlling the robotic arm to provide brushing demonstrations. This pre-trained model was then refined using DRL, optimizing the reward function to ensure effective contact pressure, full coverage of all teeth, and safe brushing mechanics. By integrating these learning algorithms, the system benefits from human expertise while reducing computational overhead and training time compared to pure reinforcement learning approaches. 

To assess the system’s performance, we introduce different evaluation metrics that measure brushing coverage, trajectory stability, and completion time. These metrics verify that the robotic arm performs brushing with consistent coverage across all teeth, smooth movements without abrupt changes, and efficient execution time to replicate a healthy person's brushing durations. Also, our results indicate that the hybrid learning approach significantly enhances performance compared to RL-based learning, demonstrating improved adaptability to different mouth and tooth structures and increased efficiency in learning optimal brushing strategies. 

This research contributes to the field of Human Robot Interaction (HRI) in assistive healthcare by showing how intelligent robotic learning techniques can be applied to autonomous dental hygiene. Our results indicate that the proposed hybrid learning framework enhances the brushing efficiency, flexibility to different dental structures and avoids overforce – important for safety and comfort in real world conditions. The effectiveness of this framework thus points to the possibility of intelligent robotic systems as assistance to individuals with disabilities, and thereby, lessen the dependence on caregivers while ensuring proper cleanliness. Thus, through the use of learning based control strategies coupled with physics based simulation environments, this work provides a starting point for further developments in robot assisted personal care and autonomous healthcare robotics. 

Presenting Author: Soroush Korivand Mississippi State University

Presenting Author Biography: Dr. Soroush Korivand is an Assistant Professor in the Michael W. Hall School of Mechanical Engineering at Mississippi State University. His research focuses on Human-Robot Interaction, AI-integrated robotics, and Reinforcement Learning, with applications in advanced manufacturing, rehabilitation, and intelligent automation. He leads interdisciplinary projects exploring adaptive control, computer vision, and biomechanical modeling. Dr. Korivand holds a Ph.D. in Mechanical Engineering and has actively contributed to national and international conferences and journals in robotics and automation.

Authors:

Parham Ahmadpanah Mississippi State University
Soroush Korivand Mississippi State University