A universal controller for any robot morphology would greatly improve computational and data efficiency. By utilizing contextual information about the properties of individual robots and exploiting their modular structure in the architecture of deep reinforcement learning agents, steps have been made towards multi-robot control. Generalization to new, unseen robots, however, remains a challenge. In this paper we hypothesize that the relevant contextual information is partially observable, but that it can be inferred through interactions for better generalization to contexts that are not seen during training. To this extent, we implement a modular recurrent architecture and evaluate its generalization performance on a large set of MuJoCo robots. The results show a substantial improved performance on robots with unseen dynamics, kinematics, and topologies, in four different environments.

Bespoken.io has developed a Test Robot that performs analog tests on voice-based hardware and software, mimicking human testing. The new robot extends Bespoken's voice testing beyond voice apps in a way that's necessary as voice technology companies and their ideas proliferate. Bespoken is best known for its virtual diagnostic tool, which is used by thousands of developers on most voice platforms to test, troubleshoot, and monitor voice apps for problems. The Test Robot goes beyond that function and into the realm of hardware by essentially recreating the way humans talk to voice apps, as can be seen in the test video at the top. The device can handle voice apps built on custom voice platforms as well as existing ones like those created by Amazon, Google, Apple, and Samsung.

The project makes FedEx the latest in a growing stream of companies to test automated, unmanned machines to make deliveries. Inc. and United Parcel Service Inc. have demonstrated drones to deliver packages in certain areas, and Amazon has displayed a rolling robot it calls Scout in trials on city streets. Grocery chain Kroger Co. recently showed off an unmanned vehicle that can deliver groceries in certain markets, and several robotics startups are testing autonomous delivery robots that use sensors and cameras to navigate sidewalks for short trips, including lunch deliveries to crowded Beijing office buildings. But on-demand delivery companies such as Deliv Inc. and DoorDash Inc. that make point-to-point trips carrying food or e-commerce purchases typically rely on armies of couriers who travel by car, scooter or bicycle. "We think this particular opportunity is going to be the most cost effective in the marketplace," FedEx Chief Marketing Officer Brie Carere said in an interview Wednesday.