Think small – maximising AI performance in industrial mobile robotics through efficiency and design simplicity

This article originally appeared in the March'24 magazine issue of Electronic Specifier Design – see ES's Magazine Archives for more featured publications.

By Amit Badlani, Director of Robotics and Generative AI, Ambarella

They work on their assigned task until told to stop; and in today’s cost-constrained world, they don’t require paying by the hour.

The robots of today are a far cry from those developed last century. As electromechanical and pneumatic technology is merged with the latest developments in AI, the latest breed of autonomous mobile robots (AMRs) are able to think for themselves, making consistently good decisions.

No longer restricted to factory production lines, modern robots are fully autonomous and able to operate anywhere they may be deployed. In applications such as ‘last mile’ deliveries, modern AMRs operate without the need for any of the guidance or preset routes that their AGV predecessors required.

The requirements for a successful AMR

However, in order to be truly practical and deployable at scale, modern AMRs need to be small in size with cost-efficient, lightweight batteries. This also means that the guidance functionality of an AMR has to be energy efficient while being powerful enough to guide the vehicle unaided. This is even more crucial in AMRs than in EVs where the battery is far larger and energy consumed by AI systems is a relatively smaller part of the total consumption.

Frugal power consumption implies less time spent charging and more time available to perform tasks, requiring fewer AMRs in total and lower capital expenditure.

Applications for AMRs are wide, diverse and growing rapidly. They include moving goods in warehouses, picking orders and ‘last mile’ delivery to help meet the increasing trend for same-day logistics. Non-payload AMRs are deployed in applications such as security where they are able to move freely around a facility, among many other applications.

Unlike earlier complex industrial robots that may well have incorporated one or more complete PCs, modern AMRs are moving to modular designs with relatively few components and subsystems. The relatively low volume production tends to preclude custom design – especially with the need to get concepts to market quickly. That is why simplicity of design is also a crucial ingredient for success; especially in a fragmented market space where new entrants with a winning idea may lack the skills or desire to begin their designs at the chip level.

Powerful energy efficient AI for robotics

With the drive to make robotic design simpler, new solutions are regularly being released to support the fast-evolving robotics market. A case in point is the new e-con Systems Robotic Computing Platform (eRCP) that merges an advanced hardware solution (based on Ambarella’s energy-efficient SoCs) with Ambarella’s designer-friendly Cooper Developer Platform, providing a solid basis for any AMR design.

Figure 1. e-con Systems’ eRCP is a modular approach to robotic design, based on Ambarella SOCs and software

Named for its emphasis on cooperation, Cooper provides a flexible and modular platform for development with any of Ambarella’s broad portfolio of AI SoCs, which are architected to provide the optimal balance between performance and power consumption. At the heart of Cooper is a software suite, known as the Cooper Foundry, which includes:

• Cooper Core: Linux-based OS, compiler and SDK
• Cooper Foundation: neural network model tools for building and deploying ML applications efficiently at the Edge
• Cooper Vision: building blocks for multi-modal sensor processing and fusion
• Cooper UX: performance analytics, sample applications and interface development tools

Figure 2. The Cooper development platform is comprehensive yet simple to use

The entire Ambarella Cooper Foundry software stack runs on Cooper Metal – a hardware layer based on the company’s AI SoCs, developer kits and modules.

Ambarella’s solutions for robotics span the whole gamut of applications, ranging from smaller robotic arms and Edge sensing modules to large last-mile delivery AMRs. With the latest generation fabricated in an advanced 5nm process technology, Ambarella’s AI SoCs lead the industry in terms of performance per watt. For example, the CV72S typically consumes less than 5W and is ideal for smaller robotics applications, whereas Ambarella’s flagship N1 SoC can be used in robotics that require more capabilities, including support for Large Language Models (LLMs). 

Summary

Robotics is truly coming of age, and we will increasingly rely on the technology in the future for a huge variety of applications, many of which we may not yet have discovered. For autonomous robots the ability to see, interpret and calculate a safe path without consuming large amounts of power is crucial.

The need for extended AMR range based upon a relatively small battery means that this advanced technology must be deployed in a small space and the advanced AI processing must be efficient. With the need to deliver new concepts to market quickly and capture the full commercial benefit, the development process must be simple, which is best achieved through modular solutions.

A perfect example of this is the eRCP based upon Ambarella’s Cooper development platform and its CV72S SoC. Barely larger than a credit card, this system-on-module (SoM) provides a complete processing solution for an advanced AMR while consuming less than 10W.