Google AI Researchers Propose TRASS Approach to Create Truly Intelligent Systems

Venturebeat | May 27, 2020

  • A preprint paper published by Stanford University and Google researchers proposes an AI technique that predicts how goals were achieved, effectively learning to reverse-engineer tasks.

  • They say it enables autonomous agents to learn through self-supervision, which some experts believe is a critical step toward truly intelligent systems.

  • A home robot could leverage it to learn tasks like turning on a computer, turning a knob, or opening a drawer, or chores like setting a dining table, making a bed, and cleaning a room.


Learning general policies for complex tasks often requires dealing with unfamiliar objects and scenes, and many methods rely on forms of supervision like expert demonstrations. But these entail significant tuning; demonstrations, for example, must be completed by experts many times over and recorded by special infrastructure.


That’s unlike the researchers’ proposed approach — time reversal as self-supervision (TRASS) — which predicts “reversed trajectories” to create sources of supervision that lead to a goal or goals. A home robot could leverage it to learn tasks like turning on a computer, turning a knob, or opening a drawer, or chores like setting a dining table, making a bed, and cleaning a room.
 

Most manipulation tasks that one would want to solve require some understanding of objects and how they interact. However, understanding object relationships in a task-specific context is non-trivial,” explain the coauthors.

Consider the task [making a bed]. Starting from a made bed, random perturbations to the bed can crumple the blanket, which when reversed provides supervision on how to flatten and spread the blanket. Similarly, randomly perturbing objects in a clean [or] organized room will distribute the objects around the room. These trajectories reversed will show objects being placed back to their correct positions, strong supervision for room cleaning


Read More: GOOGLE TO FACE ANTITRUST CHARGES FOR MONOPOLIZING ONLINE ADS


TRASS works by collecting data given a set of goal states, applying random forces to disrupt the scene, and carefully recording each of the subsequent states. A TRASS-driven agent explores outwardly using no expert knowledge, collecting a trajectory that when reversed can teach the agent to return to the goal states. In this way, TRASS essentially trains to predict the trajectories in reverse so that the trained model can take the current state as input, providing supervision toward the goal in the form of a guiding trajectory of frames (but not actions).
 

At test time, a TRASS-driven agent’s objective is to reach a state in a scene that satisfies certain specified goal conditions. At every step the trajectory is recomputed to produce a high-level guiding trajectory, and the guiding trajectory decouples high-level planning and low-level control so that it can be used as indirect supervision to produce a policy via model and model-free techniques.


In experiments, the researchers applied TRASS to the problem of configuring physical Tetris-like blocks. With a real-world robot — the Kuka IIWA — and a TRASS vision model trained in simulation and then transferred to the robot, they found that TRASS successfully paired blocks it had seen during training 75% of the time and blocks it hadn’t seen 50% of the time over the course of 20 trials each.
 

TRASS has limitations in that it can’t be applied in cases where object deformations are irreversible, for example (think cracking an egg, mixing two ingredients, or welding two parts together). But the researchers believe it can be extended by using exploration methods driven by state novelty, among other things.
 

“Our method … is able to predict unknown goal states and the trajectory to reach them,” they write. “This method used with visual model predictive control is capable of assembling Tetris-style blocks with a physical robot using only visual inputs, while using no demonstrations or explicit supervision.”

Read More: GOOGLE'S AI-POWERED ROBOT CAN GRASP OCCLUDED OBJECTS, DO NEVER-SEEN-BEFORE TASKS
 

Spotlight

Learn more about Business Value of Power Platform by enrolling into MS Power Platform Fundamentals: PL-900 Certification course.


Other News
AI TECH,SOFTWARE,FUTURE TECH

Zoom partners with Cresta to Deliver AI-Driven Real-Time Intelligence for the Contact Center

Cresta | November 10, 2022

Cresta, the leader in real-time intelligence for the contact center, today announced the availability of an integration between Cresta and Zoom Video Communications, Inc. With this integration, Zoom Contact Center customers now have the ability to deploy real-time voice integration in their contact center using Cresta's advanced AI-driven, real-time intelligence layer. This integration will allow Zoom customers to accelerate revenue generation by improving sales conversion rates, and increasing customer retention and collections. Cresta's real-time intelligence platform also improves call coaching by automating the quality assurance process. "The Zoom Contact Center solution has set a new standard for customer experiences through omnichannel and video-optimized interactions. "The Cresta integration will help Zoom Contact Center accelerate its goal of meeting the needs of the modern agent and end customer experience." Zayd Enam, CEO of Cresta Cresta's platform allows Zoom customers to use a simple interface instead of having to go through a lengthy programming process. The integration can be enabled through the Zoom administration portal, providing a simple and secure way to integrate to Cresta's powerful real-time AI suite that includes Agent Assist, Knowledge Surfacing, Supervisor coaching, QA and Insights. Customers using Cresta see CSAT increase on average by 15%, new agent ramp time decreased by 3X, and agents save 3-5 hours per week through elimination of repetitive tasks. "The Cresta platform complements the Zoom Contact Center offering," says Kentis Gopalla, Head of Ecosystem for Zoom Phone and Contact Center. "The combination of Cresta's real-time capabilities with Zoom Contact Center helps our customers improve agent productivity by providing additional insights, and deliver better customer experiences. This integration makes it easy to offer our solutions in a secure way without complex IT work, so customers can focus on business value." This partnership marks another important milestone for Cresta in 2022, including bolstering the executive team, securing an $80 million Series C funding, and the unveiling several groundbreaking products that demonstrate how AI is reshaping contact center operations, including Cresta Insights, Agent Assist, Opera, Chatbot, and Director. Cresta was also recently recognized as a "Hot AI Vendor" by Aragon Research and earned inclusion on the CB Insights AI 100 list. About Cresta Cresta makes every customer interaction excellent. Cresta turns real-time intelligence into real-time action to make the contact center smarter – and every agent and manager more productive. Powering customer experiences for companies like CarMax, Blue Nile, Earthlink, Intuit, and Porsche, Cresta is real-time AI for the real world.

Read More

AI TECH,GENERAL AI,SOFTWARE

Metatron Inc. Proudly Launches BETA Version Of Artificial Intelligence Royalty-Free Graphics Generator

Metatron, Inc. | December 01, 2022

Metatron Inc., a mobile and web technology pioneer having developed over 2,000 apps on iTunes and Google Play, is pleased to unveil the Company's first AI product in a comprehensive suite of AI products and services: "The Metatron Art Generator." Metatron is currently offering a 7-day free trial period with no credit card required, so that all interested parties may freely sign up and examine the powerful and unique new AI engine and generate an unlimited amount of royalty-free AI-generated graphics. After the trial period users will pay a reasonable monthly fee for unlimited image creation, per user. CEO Joe Riehl commented on the public release of the AI product: "The Metatron Art Generator is planned to an integral part of the "Metatron AI Suite," which will be a one-stop cloud-based solution for AI-generated digital and print content. The second product within this suite of resources is currently in ALPHA release and I'm hoping that I can unveil it to the world next week." While exploring the Metatron Art Generator users will be utilizing the most current iteration of a powerful AI engine to create any image that they can imagine! Featuring one-tap image creation, Metatron Art Generator currently creates the best and most-accurate graphic designs that modern AI can offer. Users be cautioned: entering your keywords or key-phrases and then enjoying the returned AI generated art can be addictive! Metatron thinks users will be surprised at how simple it is to generate royalty-free images for a plethora of potential uses. Below are just a few of the almost-unlimited possibilities of how people can use the graphics produced by the new Metatron Art Generator: Digital Art Metatron AI lets you work with different styles and effects, add special designs to your photos, transform images into different formats, increase or reduce different parameters so that you can create your perfect digital art. Social Media Content Metatron AI lets you create truly unique AI images that are already social media ready, and which are truly beautiful (try it!) so that you can post directly to Twitter, Facebook, Instagram, or many others. Advertising Are you advertising on social media? Metatron AI is a very powerful solution to create eye-catching images which can help boost the views and clicks of any ad campaign. Print on Demand Metatron AI can help you create just about anything you can think off just type anything and it will create any picture which is high-resolution and suitable for print media as well as digital. Do you need an image of a rabbit riding on an airplane? Or how about birds playing badminton? Watch our AI generator go to work and deliver results you thought were unimaginable without paying custom artists thousands of dollars. Graphic Design Metatron AI lets you create different types of illustrations which are totally new. And the learning curve to get wonderful results on our website takes only moments to master. Simplicity is beautiful! Photography For advanced users: Metatron AI can create incredible images effortlessly which you can use to further enhance photographs. If you're having trouble finding that "needle in a haystack" image, let Metatron Art Generator just make it for you. CEO Joe Riehl concluded: "I am so proud to announce this first in a series of AI products and services developed by Metatron. One of the things I'm most pleased about is that this software was developed totally in house, runs on Metatron's servers, and is fully owned by Metatron. Metatron will not need to share revenues with any other entities, nor did we have to give away the farm with an expensive acquisition. Furthermore, I believe that the Metatron Art Generator and subsequent AI products we release may put us on the map as a buyout candidate. One reason I believe this is because I plan to immediately begin leveraging our network of advertisers and marketers to quickly grow our subscription base."

Read More

AI TECH

Variscite Simplifies AI/ML and Multimedia at the Edge with Python API for System on Modules

Variscite | December 07, 2022

Variscite, a leading worldwide designer and manufacturer of System on Modules (SoMs), today announced the official launch of the Variscite Python API developer center. The Variscite Python API, also known as pyvar, simplifies the development of machine learning and multimedia applications for devices built on Variscite’s i.MX8-based SoMs. With the API, building and programming embedded systems and smart/edge devices for AI/ML is faster and easier, even for beginners. Variscite Python API eases the development process of embedded systems using cameras, sensors, displays, and user interfaces. It also provides an easy way to run and communicate with Cortex-M applications from the Cortex-A side, for fast processing at low power. The API’s developer center provides ‘how to’ guides, documentation and quick source code examples. “Capture, recognition, and processing of image, audio, and video data are increasingly used in embedded edge devices for any kind of environment, from transportation to healthcare, robotics, and agriculture, Our customers require additional AI/ML capabilities at the edge to run complex and advanced applications, and the Variscite python API enables faster and easier deployment with our i.MX8 SoMs.” -Ofer Austerlitz, VP Business Development and Sales of Variscite. Variscite offers a wide range of performance, power consumption, cost, and feature sets in its i.MX 8-based System on Modules. The company’s portfolio includes two highly scalable modules with dedicated AI/ML capabilities and an integrated Neural Processing Unit (NPU). The DART-MX8M-PLUS and VAR-SOM-MX8M-PLUS run on a 1.8GHz Quad Cortex™-A53 NXP iMX8M Plus processor with 800MHz Cortex-M7 real-time co-processor. Both SoMs are members of Variscite’s Pin2Pin families that offer seamless scalability, extended lifetime, as well as reduced development time, cost, and risk. ABOUT VARISCITE: Variscite is a worldwide leading System on Module designer and manufacturer, setting the bar for embedded solutions since 2003 with high-quality modules. The company provides the broadest ARM-based SoM portfolio in the embedded market with a wide range of configuration options that cover an entire embedded product and application range; from entry-level to high-performance solutions. Variscite's in-house production fully complies with the strict medical ISO13485 and ISO9001 standards. Along with the company's ongoing online documentation and personal support as well as the generous longevity, the company's customers are enjoying consistent, reliable products and services starting from the earliest development stages throughout the end-product lifecycle.

Read More

AI TECH,GENERAL AI

Aetina Launches New ASIC-Based Edge AI System Powered by Blaize

Aetina | December 08, 2022

Aetina, an edge AI solution provider that offers embedded computing hardware and software for AI and IoT, has launched a new ASIC-based edge AI system powered by the programmable Blaize® Pathfinder P1600 Embedded System on Module (SoM). The Aetina AI inference system—AIE-CP1A-A1—is a small-sized embedded computer designed for different computer vision applications, including object detection, human motion detection, and automated inspection. The AIE-CP1A-A1 model features the small form factor Blaize® Pathfinder P1600 Embedded SoM, ideal for rugged and challenging environments and embedded systems that need extended temperature ranges. It features Dual ARM Cortex A53 processors, 16 GSP cores providing 16 TOPS AI performance, and the benefits of low memory bandwidth and low latency, making it capable of accelerating neural network (NN) deployment and deep learning (DL) process. Based on the Blaize Graph Streaming Processor (GSP®) architecture, the Pathfinder enables energy-efficient processing power ideal for AI inferencing workloads at the edge, easily plugs into a custom carrier board, and runs autonomously or integrated within a larger system. Due to its compact, fanless design and multiple I/O ports, the AIE-CP1A-A1 can fit both commercial and industrial-grade AI-powered systems that involve vision AI-related inference. The Blaize Pathfinder also supports H.264/H.265 video encoder and decoder, making it compatible with different sensors to run quick image recognition and video analytics tasks. Aetina offers customized carrier boards and computer cases for sophisticated system integration and chip-down design service to provide other kinds of embedded computing models, such as single-board computer (SBC) with the Blaize Pathfinder in addition to regular box PCs. Moreover, Aetina helps developers convert their AI inference models to a format that can run and function normally on edge devices built with ASIC chips. “We are now expanding our edge computing product line to bring more GPU and ASIC-based solutions to help developers build their AI systems in various verticals and industries. “The AIE-CP1A-A1 model, powered by Blaize Pathfinder P1600 SoM, is our first ASIC-based system. In the future, by integrating different types of chips and modules, we can offer heterogeneous computing devices that are suitable for different AI models.” Jackal Chen, Senior Product Manager at Aetina Keeping AI processing and inferencing workloads at the edge, rather than sending data to the cloud, is critical for cost-effective and almost latency-free AI applications," said Dinakar Munagala, CEO and Co-founder of Blaize. "The Pathfinder SoM and our GSP architecture enable this key business objective for the demands of computer vision applications, including object detection, autonomous optical inspection, robotics, and more." Additionally, Aetina is now developing more ASIC-based AI accelerators with M.2 and EDSFF specification for developers, speeding up the inference processes of their AI models at the edge. About Blaize Blaize is a leading provider of a proprietary purpose-built, full-stack hardware architecture and low-code/no-code software platform that enables edge AI processing solutions at the network’s edge for computing in multiple large and rapidly-growing markets — automotive, mobility, retail, security, industrial automation, medical devices, and many others. Blaize’s novel solution solves the technical problem that edge AI processing requires across those verticals — very low latency and high thermal and power efficiency — which previously relied on retrofitting sub-optimized AI solutions designed more for data centers and the cloud. Blaize has previously raised over $180MM from strategic investors such as DENSO, Daimler, Magna, and Samsung, and financial investors such as Franklin Templeton, Temasek, GGV, and others. With headquarters in El Dorado Hills (CA), Blaize has teams in San Jose (CA) and subsidiaries in Hyderabad (India), Manila (Philippines), and Leeds and Kings Langley (UK), with 300+ employees worldwide. About Aetina Aetina Corporation, founded in 2012 in Taipei, Taiwan, is a leading provider of state-of-the-art AI solutions involving high-performance edge computing hardware and low-code/no-code software to help developers build different AI verticals quickly. With a focus on the global AI market, Aetina is committed in delivering professional longevity services to solve the developers’ pain points and speed up their AI adoption processes.

Read More