Research in Purdue SMART Lab

Robot learning is a field that enables robots to acquire and improve their capabilities through various learning techniques and algorithms. At the SMART Lab, we push the boundaries of robot intelligence through several interconnected research directions. We develop advanced learning algorithms that enhance robots' ability to perceive, reason about, and interact with their environment. Another significant portion of our research explores reinforcement learning approaches that allow robots to learn optimal behaviors through experience and interaction. We are particularly interested in human-in-the-loop robot learning, including Learning from Demonstration (LfD) and preference learning, which enables robots to learn directly from human teachers and adapt to human preferences. Recently, we have begun investigating how generative AI technologies can revolutionize robot learning, exploring ways to leverage large language models and other generative AI systems to enhance robots' learning capabilities and decision-making processes. This cutting-edge research aims to create more adaptable and intelligent robotic systems that can better understand and respond to human needs.

You can learn more about our current and past research on robot learning below.

 

Description: Generative AI (GAI), particularly Large Language Models (LLMs) and Vision-Language Models (VLMs), represents a groundbreaking advancement in artificial intelligence. These foundation models, trained on vast datasets of knowledge and experience, exhibit remarkable general reasoning capabilities that can be harnessed for robotics applications. At the SMART Lab, we are at the forefront of integrating generative AI with robotic systems to enhance their cognitive and learning abilities. Our recent research explores multiple innovative applications: utilizing LLMs for intelligent task allocation and coordination in multi-robot and human-robot teams; developing LLM-powered approaches for improved robot semantic navigation and scene understanding; and investigating how crowdsourced LLMs can serve as synthetic teachers in robot learning scenarios, providing valuable feedback and guidance. By harnessing the power of generative AI, we aim to revolutionize the way robots reason, learn, and interact, paving the way for more capable and adaptable robotic systems.

Grants: NSF, Purdue University
People: Vishnunandan Venkatesh, Ruiqi Wang, Taehyeon Kim, Ziqin Yuan, Ikechukwu Obi, Arjun Gupte

Selected Publications:

• Shyam Sundar Kannan*, L. N. Vishnunandan Venkatesh*, and Byung-Cheol Min (*equal contribution), "SMARTLLM: Smart Multi-Agent Robot Task Planning using Large Language Models", 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Abu Dhabi, UAE, October 13-17, 2024. Paper Link, Video Link
• Taehyeon Kim and Byung-Cheol Min, "Semantic Layering in Room Segmentation via LLMs", 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Abu Dhabi, UAE, October 13-17, 2024. Paper Link, Video Link
• Ruiqi Wang*, Dezhong Zhao*, Ziqin Yuan, Ike Obi, and Byung-Cheol Min (* equal contribution), "PrefCLM: Enhancing Preference-based Reinforcement Learning with Crowdsourced Large Language Models", IEEE Robotics and Automation Letters. (Under Review) Paper Link, Video Link

 

Description: Reinforcement Learning (RL) traditionally requires precisely defined reward functions to guide robot behavior, a challenging requirement in complex human-robot interaction scenarios. Preference-based Reinforcement Learning (PbRL) offers an innovative solution by learning from human comparative feedback rather than predefined rewards, making it more intuitive to teach robots desired behaviors. At the SMART Lab, we are advancing PbRL techniques to address key challenges in the field, such as the need for efficient learning from minimal human feedback and the complexity of modeling human preferences. Our research focuses on developing algorithms that can better interpret various forms of human feedback while requiring fewer interactions. Through innovations like our feedback-efficient active preference learning approach, we aim to make robot learning more natural and practical for real-world applications.

Grants: NSF, Purdue University
People: Ruiqi Wang, Weizheng Wang, Ziqin Yuan, Ikechukwu Obi

Selected Publications:

• Weizheng Wang, Ruiqi Wang, Le Mao, and Byung-Cheol Min, "NaviSTAR: Benchmarking Socially Aware Robot Navigation with Hybrid Spatio-Temporal Graph Transformer and Active Learning", 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2023), Detroit, USA, October 1-5, 2023. Paper Link, Video Link, GitHub Link
• Ruiqi Wang, Weizheng Wang, and Byung-Cheol Min, "Feedback-efficient Active Preference Learning for Socially Aware Robot Navigation", 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2022), Kyoto, Japan, October 23-27, 2022. Paper Link, Video Link, GitHub Link

Description: Learning from Demonstration (LfD) is a powerful paradigm that enables robots to learn new skills by observing and replicating human demonstrations. This approach bridges the gap between human expertise and robot capabilities, making it more intuitive for non-experts to teach robots complex behaviors. At the SMART Lab, we are advancing the frontiers of LfD research, with a particular focus on its application to Multi-Robot Systems (MRS). While traditional LfD has primarily focused on single-robot scenarios, we are pioneering methods to extend these principles to multiple robots working in coordination. Our innovative framework leverages visual demonstrations to capture intricate robot-object interactions and complex collaborative behaviors. By developing sophisticated algorithms that can translate human demonstrations into coordinated multi-robot actions, we aim to make robot teaching more accessible and efficient.

Grants: NSF, Purdue University
People: Vishnunandan Venkatesh, Taehyeon Kim, Ruiqi Wang

Selected Publications:

• L. N. Vishnunandan Venkatesh and Byung-Cheol Min, "Learning from Demonstration Framework for Multi-Robot Systems Using Interaction Keypoints and Soft Actor-Critic Methods", 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Abu Dhabi, UAE, October 13-17, 2024. (Paper Link, Video Link)
• Ruiqi Wang, Weizheng Wang, and Byung-Cheol Min, "Feedback-efficient Active Preference Learning for Socially Aware Robot Navigation", 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2022), Kyoto, Japan, October 23-27, 2022. Paper Link, Video Link, GitHub Link

Description: Visual localization enables autonomous vehicles and robots to navigate based on visual observations of their operating environment. In visual localization, the agent estimates its pose based on the image from the camera. The operating environment of the agent can undergo various changes due to illumination, day and night, seasons, structural changes, and so on. In vision-based localization, it is important to adapt to these changes that can significantly impact visual perception. The SMART lab investigates into developing methods that enable autonomous agents to robustly localize despite these changes in the surroundings. For example, we developed a visual place recognition system that aids the autonomous agent in identifying its location on a large-scale map by retrieving a reference image that matches closely with the query image from the camera. The prposed method utilizes consice descriptors from the image, so that the image process can be done rapidly with less memory consumption.

Grants: NSF, Purdue University
People: Shyam Sundar Kannan, Vishnunandan Venkatesh

Selected Publications:

• Shyam Sundar Kannan and Byung-Cheol Min, "PlaceFormer: Transformer-based Visual Place Recognition using Multi-Scale Patch Selection and Fusion", IEEE Robotics and Automation Letters, Vol. 9, No. 7, pp. 6552-6559, July 2024. Paper Link
  

Learning-based Robot Recognition (2017 - 2022)

Application Offloading Problem (2018 - 2022)