RoboCraft: Learning to see, simulate, and shape elasto-plastic objects in 3D with graph networks

Article Robotics

Sample-Efficient Learning of Deformable Linear Object Manipulation in the Real World Through Self-Supervision

Robert Lee et al.

Summary: In this study, a novel learning approach for deformable object manipulation is proposed, which achieves goal reaching tasks with only a small amount of real-world data.

IEEE ROBOTICS AND AUTOMATION LETTERS (2022)

Add to Collection

Article Computer Science, Artificial Intelligence

VisuoSpatial Foresight for physical sequential fabric manipulation

Ryan Hoque et al.

Summary: Robotic fabric manipulation has applications in various fields, but existing techniques are often task-specific and hard to generalize. Researchers leverage the Visual Foresight framework to efficiently reuse learned fabric dynamics for different sequential fabric manipulation tasks.

AUTONOMOUS ROBOTS (2022)

Add to Collection

Article Robotics

A Bayesian Treatment of Real-to-Sim for Deformable Object Manipulation

Rika Antonova et al.

Summary: In this research, the problem of inferring simulation parameters for deformable objects is addressed. By defining the state space of an object and embedding noisy keypoints extracted from images into a distribution, the motion of the object in different images can be represented as a trajectory of distribution embeddings in the deformable state space. This approach allows for incorporating noisy state observations into Bayesian inference tools and estimating posterior distributions over simulation parameters for highly deformable objects.

IEEE ROBOTICS AND AUTOMATION LETTERS (2022)

Add to Collection

Proceedings Paper Automation & Control Systems

SpeedFolding: Learning Efficient Bimanual Folding of Garments

Yahav Avigal et al.

Summary: This research develops a reliable and efficient bimanual system called SpeedFolding, which utilizes a novel neural network architecture to predict gripper poses and manipulate crumpled garments into smooth and folded configurations. Experimental results demonstrate the system's ability to generalize to different garments and significantly increase folding speed.

2022 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS (IROS) (2022)

Add to Collection

Article Robotics

Differentiable Simulation for Physical System Identification

Quentin Le Lidec et al.

Summary: The research proposes a more accurate algorithm for solving the problem of frictional contacts and introduces a differentiable simulator as well as an efficient method for computing the analytical derivatives of the optimization problems involved.

IEEE ROBOTICS AND AUTOMATION LETTERS (2021)

Add to Collection

Article Robotics

Cable manipulation with a tactile-reactive gripper

Yu She et al.

Summary: This paper presents a novel perception and control framework for real-time cable following using robotic grippers equipped with vision-based tactile sensors. By combining two tactile-based controllers, the system successfully achieves cable gripping and alignment, preventing the cable from slipping out of the gripper.

INTERNATIONAL JOURNAL OF ROBOTICS RESEARCH (2021)

Add to Collection

Review Robotics

Modeling, learning, perception, and control methods for deformable object manipulation

Hang Yin et al.

Summary: The importance of perceiving and handling deformable objects in our daily life is highlighted in this paper, with a focus on automating tasks like food handling and garment sorting. Advances in data-driven approaches, combined with traditional control and planning methods, offer viable solutions to these challenges. A learning perspective is utilized to unify discussions on analytical and data-driven methods, addressing the integration of model priors and task data in manipulating deformable objects.

SCIENCE ROBOTICS (2021)

Add to Collection

Proceedings Paper Automation & Control Systems

Learning to Rearrange Deformable Cables, Fabrics, and Bags with Goal-Conditioned Transporter Networks

Daniel Seita et al.

Summary: Manipulating deformable objects such as cables, fabrics, and bags poses challenges due to their complex dynamics and high-dimensional configuration spaces, making goal specification and multi-step planning difficult.

2021 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA 2021) (2021)

Add to Collection

Proceedings Paper Automation & Control Systems

Deformable Elasto-Plastic Object Shaping using an Elastic Hand and Model-Based Reinforcement Learning

Carolyn Matl et al.

Summary: This study proposes a method of shaping elasto-plastic dough using a novel elastic end-effector in a reinforcement learning framework. Experimental results show that the proposed framework requires 60% fewer actions than a heuristic method in rolling dough to a specified length. Furthermore, estimating stiffness using the soft end-effector can effectively initialize models, improving robot performance by approximately 40%.

2021 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS (IROS) (2021)

Add to Collection

Article Computer Science, Artificial Intelligence

Automatic tuning of hyperparameters using Bayesian optimization

A. Helen Victoria et al.

Summary: Deep learning in artificial intelligence is effective in computer vision, natural language processing, and audio recognition. Hyperparameter tuning is crucial for training model performance, and Bayesian optimization algorithm enhances model performance effectively.

EVOLVING SYSTEMS (2021)

Add to Collection

Article Robotics