Multi-agent path finding (MAPF) is an essential component of many large-scale, real-world robot deployments, from aerial swarms to warehouse automation. However, despite the community's continued efforts, most state-of-the-art MAPF planners still rely on centralized planning and scale poorly past a few hundred agents. Such planning approaches are maladapted to real-world deployments, where noise and uncertainty often require paths be recomputed online, which is impossible when planning times are in seconds to minutes. We present PRIMAL, a novel framework for MAPF that combines reinforcement and imitation learning to teach fully-decentralized policies, where agents reactively plan paths online in a partially-observable world while exhibiting implicit coordination. This framework extends our previous work on distributed learning of collaborative policies by introducing demonstrations of an expert MAPF planner during training, as well as careful reward shaping and environment sampling. Once learned, the resulting policy can be copied onto any number of agents and naturally scales to different team sizes and world dimensions. We present results on randomized worlds with up to 1024 agents and compare success rates against state-of-theart MAPF planners. Finally, we experimentally validate the learned policies in a hybrid simulation of a factory mockup, involving both real-world and simulated robots.

Dexterous hands have the ability to grasp arbitrary parts and impart arbitrary motions to those objects. They also, in general, can be placed in special configurations where the grasp of an object is underconstrained or overconstrained.It is also possible to obtain special configurations where there are not enough or there are redundant degrees of freedom for moving an object, or where the hand mechanism is in a singular Configuration. All of these special configurations are revealed through the examination of linear Jacobian matrix relationships describing a grasp. Formulations are presented for determining what useful manipulations can be performed, given these circumstances.

Foaming of stainless steelmaking slags is difficult because the gas-generation rate is normally low. In the present work, the impact of reducible oxides (such as NiO by carbon), waste oxide briquettes, limestone, and calcium nitrate on foaming was examined. Specifically, the gas generation rates and the rate-controlling mechanisms for each of these additives were determined. From these results, the expected foaming in actual steelmaking operations can be predicted. The NiO and carbon additions do not appear practical for generating foam. However, limestone, calcium nitrate, and possibly waste oxide briquettes should provide significant foaming.

The ability to grasp and manipulate transparent objects is a major challenge for robots. Existing depth cameras have difficulty detecting, localizing, and inferring the geometry of such objects. We propose using neural radiance fields (NeRF) to detect, localize, and infer the geometry of transparent objects with sufficient accuracy to find and grasp them securely. We leverage NeRF's viewindependent learned density, place lights to increase specular reflections, and perform a transparency-aware depth-rendering that we feed into the Dex-Net grasp planner. We show how additional lights create specular reflections that improve the quality of the depth map, and test a setup for a robot workcell equipped with an array of cameras to perform transparent object manipulation. We also create synthetic and real datasets of transparent objects in real-world settings, including singulated objects, cluttered tables, and the top rack of a dishwasher. In each setting we show that NeRF and Dex-Net are able to reliably compute robust grasps on transparent objects, achieving 90 % and 100 % grasp success rates in physical experiments on an ABB YuMi, on objects where baseline methods fail. See https://sites.google.com/view/dex-nerf for code, video, and datasets.