LOVE PANTA

Mr. LOVE PANTA


Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications

11 March 2024 in Research Square
Verified
Preprint
Open access
PDF file
LPLOVE PANTA
LOVE PANTA
Trajectory tracking for an Omni-drive robot presents a challenging task that demands an efficient controller design. To address the limitations of manual tuning, we introduce a self-optimizing controller named fuzzyPID, leveraging the analysis of responses from various dynamic and static systems. The rule-based controller design is implemented using Matlab/Simulink, and trajectory tracking simulations are conducted within the CoppeliaSim environment. Similarly, a non-linear model predictive controller(NMPC) is proposed to compare tracking performance with fuzzyPID. We also assess the impact of tunable parameters of NMPC on its tracking accuracy. Simulation results validate the precision and effectiveness of NMPC over fuzzyPID controller while trading computational complexity.

20 October 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
LPLOVE PANTA
LOVE PANTA
Trajectory tracking for holonomic drive systems remains a significant challenge, particularly in complex and dynamic environments. This demands the development of effective control algorithms capable of following pre-defined trajectories while simultaneously avoiding obstacles. In this work, we propose a model-free deep reinforcement learning (DRL) approach integrated with curriculum learning for omni-directional mobile robots. We design a specialized RL architecture that leverages only the most relevant inputs, such as trajectory error and relative obstacle positions, to enable efficient and safe path planning. The control policies are trained in a simulated environment using a diverse trajectory patterns and external disturbances. Curriculum learning is applied hierarchically, gradually increasing environmental complexity to enhance both learning efficiency and generalization capability. The performance of the tracking model is quantitatively evaluated using the success rate and mean tracking error. Simulation results demonstrate a 90 percent success rate across 100 randomized tracking tasks in previously unseen complex scenarios. Furthermore, the proposed DRL controller with curriculum learning exhibits improved stability, faster training convergence, and robust disturbance rejection capabilities compared to learning from scratch.