A robot autonomously completes tasks by transforming its body to match capabilities newly encountered environments.

We present the methodology, algorithms, system design, and experiments addressing self-assembly of large teams autonomous robotic boats into floating platforms. Identical self-propelled autonomously dock together form connected structures with controllable variable stiffness. These can self-reconfigure arbitrary shapes limited only by number rectangular elements assembled in brick-like patterns. An O(m <sup xmlns:mml="http://www.w3.org/1998/Math/MathML"...

We present the EP-Face connector, a novel connector for hybrid chain-lattice type modular robots that is highstrength (88.4N), compact, fast, power efficient, and robust to position errors. The consists of an array electro-permanent magnets (EP magnets) embedded in planar face. EP are solid-state can be turned on off require only when changing state. In this paper, we design manufacturing process, detailed experimental characterization strength under different loading conditions, compare its...

This paper addresses the self-assembly of a large team autonomous boats into floating platforms. We describe design individual boats, systems concept, algorithms, software architecture and experimental results with prototypes that are 1:12 scale realizations modified ISO shipping containers, goal demonstrating maritime structures such as air strips, bridges, harbors or sea bases. Each container is robotic module capable holonomic motion can dock in brick pattern to form arbitrary shapes....

The advantage of modular robot systems lies in their flexibility, but this can only be realized if there exists some reliable, effective way generating configurations (shapes) and behaviors (controlling programs) appropriate for a given task.In paper, we present an end-to-end system addressing tasks with robots, demonstrate that it is capable accomplishing challenging multi-part hardware experiments.The consists four tightly integrated components: (1) A high-level mission planner, (2) large...

We present a novel approach to robotic grasp planning using both learned proposal network and 3D shape reconstruction network. Our system generates 6-DOF grasps from single RGB-D image of the target object, which is provided as input networks. By geometric refine candidate produced by network, our able accurately known unknown objects, even when location on object not visible in image.This paper presents architectures, training procedures, refinement method that comprise system. Experiments...

The advantage of modular robot systems lies in their flexibility, but this can only be realized if there exists some reliable, effective way generating configurations (shapes) and behaviors (controlling programs) appropriate for a given task. In paper, we present an end-to-end system addressing tasks with robots, demonstrate that it is capable accomplishing challenging multi-part hardware experiments. consists four tightly integrated components: (1) A high-level mission planner, (2) design...

This paper presents a method for kinematic retargeting that is general to broad class of chains. Kinematic the adaptation pose or motion from one embodiment another. Our distinguishes itself in its ability adapt poses new robots with very little configuration by user. We accomplish this defining two metrics and minimizing cost function which weighted sum these metrics. allows automatically between source target chains have different link lengths degrees freedom. These capabilities address...

We present a system enabling modular robot to autonomously build structures in order accomplish high-level tasks. Building allows the surmount large obstacles, expanding set of tasks it can perform. This addresses common weakness systems, which often struggle traverse obstacles. paper presents hardware, perception, and planning tools that comprise our system. An environment characterization algorithm identifies features be augmented create path between two disconnected regions environment....

Abstract Accurate position sensing is important for state estimation and control in robotics. Reliable accurate sensors are usually expensive difficult to customize. Incorporating them into systems that have very tight volume constraints such as modular robots particularly difficult. PaintPots low-cost, reliable, highly customizable sensors, but their performance dependent on the manufacturing calibration process. This paper presents a Kalman filter with simplified observation model...

The PaintPot manufacturing process is a new way to create low-cost, low-profile, highly customizable potentiometers for position sensing in robotic applications. It uses widely accessible materials, requires no special expertise, and creates custom variety of shapes sizes, including curved surfaces. PaintPots offer accuracy precision performance comparable with commercial (non-customizable) options through calibration that trades small computation cost. This paper includes detailed...

We address the problem of detecting embeddability modular robots: namely, to decide automatically whether a given robot design can simulate functionality seemingly different design. To that end, we introduce novel graph representation for robots and formalize notion embedding through topological kinematic conditions. Based on that, develop an algorithm decides when two involved designs have tree topologies. Our performs passes involves dynamic programming maximum cardinality matching....

We present a novel method enabling robots to quickly learn manipulate objects by leveraging motion planner generate "expert" training trajectories from small amount of human-labeled data. In contrast the traditional sense-plan-act cycle, we propose deep learning architecture and regimen called PtPNet that can estimate effective end-effector for manipulation directly single RGB-D image an object. Additionally, data collection augmentation pipeline enables automatic generation large numbers...

uilding structures can allow a robot to surmount large obstacles, expanding the set of areas it reach. This paper presents planning algorithm automatically determine what construction-capable must build in order traverse its entire environment. Given an environment, building blocks, and capable structures, we seek optimal (using minimum number blocks) that could be built make environment traversable with respect robot's movement capabilities. We show this problem is NP-Hard, present...

We present a novel approach to robotic grasp planning using both learned proposal network and 3D shape reconstruction network. Our system generates 6-DOF grasps from single RGB-D image of the target object, which is provided as input networks. By geometric refine candidate produced by network, our able accurately known unknown objects, even when location on object not visible in image. This paper presents architectures, training procedures, refinement method that comprise system. Experiments...

We present a system enabling modular robot to autonomously build structures in order accomplish high-level tasks. Building allows the surmount large obstacles, expanding set of tasks it can perform. This addresses common weakness systems, which often struggle traverse obstacles. paper presents hardware, perception, and planning tools that comprise our system. An environment characterization algorithm identifies features be augmented create path between two disconnected regions environment....

Building structures can allow a robot to surmount large obstacles, expanding the set of areas it reach. This paper presents planning algorithm automatically determine what construction-capable must build in order traverse its entire environment. Given an environment, building blocks, and capable structures, we seek optimal (using minimum number blocks) that could be built make environment traversable with respect robot's movement capabilities. We show this problem is NP-Hard, present...