Publications
Peer-reviewed work from the PhD, most of it on getting learned control to survive contact with real terrain.
Deep Reinforcement Learning for Coordinated Payload Transport in Biped-Wheeled Robots
IEEE International Conference on Robotics and Automation (ICRA), pp. 14992–14998
A single deep reinforcement learning agent controls two biped-wheeled robots carrying a shared payload, trained with massively parallel simulation in Isaac Lab and transferred to hardware zero-shot within a 0.5 m error range.
@inproceedings{mehta2025payload,
author = {Mehta, Dhruv and Joglekar, A. and Krovi, Venkat},
title = {Deep Reinforcement Learning for Coordinated Payload Transport in Biped-Wheeled Robots},
booktitle = {IEEE International Conference on Robotics and Automation (ICRA)},
pages = {14992--14998},
year = {2025}
}
Agile Off-Road Terrain Traversal of an Ackermann Steered Platform using Deep Reinforcement Learning
IFAC-PapersOnLine (Modeling, Estimation and Control Conference), Vol. 59(3), pp. 79–84
Learning-based goal-to-goal navigation with a point stabilization task for a mid-scale off-road platform, with reward shaping tuned for stability and a custom off-road simulation environment. 86% success rate at reaching the desired goal.
@article{mehta2025offroad,
author = {Mehta, Dhruv and Salvi, Ameya and Krovi, Venkat},
title = {Agile Off-Road Terrain Traversal of an Ackermann Steered Platform using Deep Reinforcement Learning},
journal = {IFAC-PapersOnLine},
volume = {59},
number = {3},
pages = {79--84},
year = {2025},
note = {Modeling, Estimation and Control Conference (MECC)}
}
Rough Terrain Path Tracking of an Ackermann Steered Platform using Hybrid Deep Reinforcement Learning
IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM), pp. 685–690
A model-based controller combined with a deep reinforcement learning policy for path tracking across diverse terrains, with rough-terrain assets authored in Blender and exported to OpenUSD. 98.5% precision on unseen real-world tracks.
@inproceedings{mehta2024roughterrain,
author = {Mehta, Dhruv and Salvi, Ameya and Krovi, Venkat},
title = {Rough Terrain Path Tracking of an Ackermann Steered Platform using Hybrid Deep Reinforcement Learning},
booktitle = {IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM)},
pages = {685--690},
year = {2024}
}
Actively Articulated Wheeled Architectures for Autonomous Ground Vehicles – Opportunities and Challenges
SAE Technical Paper 2023-01-0109
A survey of actively articulated wheeled architectures for autonomous ground vehicles: where the added mechanical degrees of freedom buy real capability off-road, and what they cost in control complexity. This review framed the direction of my dissertation work.
@techreport{mehta2023articulated,
author = {Mehta, Dhruv and Kosaraju, K. C. and Krovi, Venkat},
title = {Actively Articulated Wheeled Architectures for Autonomous Ground Vehicles -- Opportunities and Challenges},
institution = {SAE International},
number = {2023-01-0109},
type = {SAE Technical Paper},
year = {2023}
}
Vibration Control in Meta-Structures Using Reinforcement Learning
Conference Proceedings of the Society for Experimental Mechanics Series, Springer
Reinforcement learning applied to active vibration control in meta-structures — my first published use of RL, from before the robotics work, and where the interest in learned control started.
@incollection{mehta2022vibration,
author = {Mehta, Dhruv and Malladi, S.},
title = {Vibration Control in Meta-Structures Using Reinforcement Learning},
booktitle = {Conference Proceedings of the Society for Experimental Mechanics Series},
publisher = {Springer},
year = {2022}
}