We propose a Deep Interaction Prediction Network (DIPN) for learning to predict complex interactions that ensue as a robot end-effector pushes multiple objects, whose physical properties, including size, shape, mass, and friction coefficients may be unknown a priori. DIPN "imagines" the effect of a push action and generates an accurate synthetic image of the predicted outcome. DIPN is shown to be sample efficient when trained in simulation or with a real robotic system. The high accuracy of DIPN allows direct integration with a grasp network, yielding a robotic manipulation system capable of executing challenging clutter removal tasks while being trained in a fully self-supervised manner. The overall network demonstrates intelligent behavior in selecting proper actions between push and grasp for completing clutter removal tasks and significantly outperforms the previous state-of-the-art. Remarkably, DIPN achieves even better performance on the real robotic hardware system than in simulation.

Computer Vision and Robotic System for Recycling Automation - We are working with a large recycling company to help it modernizing its product lines to increase levels of automation as an early prototype, we build a demo system for the separation of scrap metals based on color that is robust to lighting condition changes. The video shows a real-time demo of a run that separates pure copper pieces from mixed aluminum-copper pieces. We expect to report more exciting results as the project progresses.

DDM - DDM solves one-shot and dynamic optimal multi-robot path planning problems in a graph-based setting. DDM is mainly enabled through exploiting two innovative heuristics: path diversification and optimal sub-problem solution databases. The two heuristics attack two distinct phases of a decoupling-based planning planner: while path diversification allows more effective use of the entire workspace for robot travel, optimal sub-problem solution databases facilitate the fast resolution of local path conflicts.

Efficient High Quality Stack Rearrangement - Video highlight of our RA-L/ICRA 2018 work with the same name. Abstract: We study a variant of rearrangement problems that appear frequently in applications, which involves sorting objects or robots in stack-like containers that can be accessed only from one side. We provide efficient algorithms that could generate high quality rearrangement sequence.

Optimal Tabletop Object Rearrangement with Overhand Grasps - Video highlight of our RSS 2017 work on optimal tabletop object rearrangement and subsequent extended version. Our hardware experiments confirm our hypothesis that (1) grasping/releasing is generally much more time consuming and (2) our proposed algorithm provide significant benefit when compared with a greedy algorithm.

A Portable, 3D-Printing Enabled Multi-Vehicle Platform for Robotics Research and Education - Video highlight of our microMVP platform for all! See https://arc.cs.rutgers.edu/mvp/ for more details or read more here.

Near-Optimal Multi-Robot Path Planning in Continuous Domain - Video highlights accompanying our ISRR work. You may also [download the video].

Optimal Reconfiguration of Multi-Robot Formations - In two (CDC'12, CDC'13) recent works, we developed efficient algorithm for the distance optimal reconfiguration of multi-robot formations. The video below demonstrates effectiveness of the algorithm. We note that the examples in the video take less than 0.1 second to solve when implemented in Java and running on a MacBook Air (2013). [download the video].

Rendezvous Without Coordinates - This research establishes a sufficient condition for an arbitrary (known) number of Dubins-car vehicles to rendezvous in finite time. The sensing model of the vehicle is extremely coarse with only three quantized values. The feedback control law is similarly quantized with three total control input. In particular, the vehicles do not perform any state estimation, i.e., no coordinate data is needed. Our result generalizes to distributed systems without central coordination as well as in-homogeneous vehicles.
The video below demonstrates the sufficient condition for rendezvous, which depends solely on the sensor quantization (windshield size). We show two cases of rendezvous and two cases of divergence. Time evolutions of both the system and the Lyapunov certificate are shown. The simulation program is fully accessible here. [download the video].