How Amazon Astro moves smoothly through its environment

Amazon recently gave more information about how Astro, the company’s multip-purpose home robot, is able to move smoothly through its environment with limited onboard computational capabilities. 

Astro’s sensor field of view and onboard computational capabilities aren’t nearly as powerful as other autonomous robots, like autonomous vehicles. This makes it a much more affordable option for consumers, but it also means it’s more challenging for Amazon to deliver a high-quality of motion. 

Amazon counteracts Astro’s lack of computation capabilities with algorithms and software designed to allow the robot to move more gracefully. 

Predictive planning is a key aspect of Astro’s navigational abilities. Astro’s limited computational capabilities mean it struggles with a large sensing-to-actuation latency. To combat this, Astro makes predictions about the movements of the objects around it, like people. The robot predicts where those objects will be and what its surroundings will look like at the end of its current planning cycle, helping it to account for latencies in sensing and mapping while it’s moving.

All of Astro’s plans are based on its latest sensor data and what it thinks its surroundings will look like when its plan will be taking effect. The robot can make these predictions because of its ability to predict and handle uncertainties and risks of collisions. 

Astro’s motivation to move towards its goal is always weighed dynamically with its perceived level of uncertainty. This means that Astro evaluates uncertainty-adjusted progress for each candidate motion, allowing it to focus on getting to its goal when it determines risk is low, and focus on evasion when risk is high. 

The robot also utilizes trajectory optimization software to operate smoothly in its environment. Astro considers multiple candidate trajectories and picks the best one in each planning cycle. The robot plans 10 times a second and evaluates a few hundred trajectory candidates in each instance. 

Astro considers safety, smoothness of motion and progress toward its end goal. With these three criteria, the robot picks the trajectory that will result in optimal behavior. Other approaches limit the number of choices a robot can make to a discrete set, or a state lattice, but Amazon’s formulation is continuous, helping the robot move smoothly. 

Astro doesn’t just have to plan where its two wheels and body will go, it also has to plan movements for Astro’s screen. The robot’s screen is used to communicate motion and intent and for active perception, so Astro plans to do things like orienting its screen towards the person it’s following or in the direction it plans to go so humans around it know what its plans are. 

Amazon released Astro in September 2021. The robot can be used for a variety of things, including home monitoring, videoconferencing with family and friends, entertaining children, and more. The voice-controllable robot can recognize faces, deliver items to specific people, after a human puts the item in the storage bin, and use third-party accessories to, for example, record blood pressure. It can detect the sound of a smoke alarm, carbon monoxide detector or breaking glass. If you have a Ring account, Astro can send you notifications if it notices something unusual.

The post How Amazon Astro moves smoothly through its environment appeared first on The Robot Report.

from The Robot Report - Robotics News, Analysis & Research https://ift.tt/pxLh8FN
via artificialconference