A telepresence robot is a machine that interacts with humans to help them to interact with people and their surrounding environment remotely. These robots enable virtual visits at hospitals and nursing homes, enable virtual walk-throughs for home buyers located far away from a property, and keep students connected with their classwork and peers when they are out sick.
A telepresence robot is, effectively, HMI on steroids.
We caught up with Jared Go, CTO at OhmniLabs, to get a description of how to build a better telepresence robot.
Design News: What are the qualities that are critical to building a telepresence robot?
Jared Go: A telepresence robot seems like a simple thing, right? You have a video chat, like any tablet, and it moves around. But when it comes to building one, you'll find quickly, there are so many things that go into it. You have to design a power system and optimize power to have these robots last several hours between charges. People rarely use these robots for a quick call, they usually log in to them and remain on them for three to five hours. With that comes designing a battery subsystem with lithium-ion batteries, and with lithium-ion batteries comes safety, and with that comes so many other things - and again, that’s just for the batteries. And then because it's a robot and someone else is remotely controlling it, you need a way to charge it, without requiring somebody to physically plug it in after they’re done using it - thus you have to develop some kind of docking system and build the robot to navigate itself to the docking station safely so that it can charge.
You also need the charger cable to supply sufficient power for quick charging, so that the robot is always ready to go. Additionally, you need it to be safe for a multitude of environments, including homes, hospitals, office buildings, nursing homes, and more, so it has to be non-hazardous to children, the elderly, and pets, and you need to make it non-obtrusive, if you have a massive charger and you're trying to deploy this in a certain environment, let's say a cramped hospital room, it's a no go. So, there are all these small factors to consider in design.
You have to be an expert at all sorts of layers. So, there's, of course, the mechanical layers, there are the electrical layers, and there are the software layers. On the mechanical side, there are simple things to engineer like putting a screen and camera up on a long, tall robot, but in doing this, you get movement with every dip or crack in the floor, right? You also get flex in the tube; you get flex in the system. And so, the challenge is in designing the system to respect the dynamics of everything you've created. And you can do that in different ways. You can have firmware that makes your acceleration smooth, you can design the mechanics to be stiffer or to absorb the shock, you can have software that does image stabilization. There are so many ways to solve these problems, but it requires a team that's capable of looking across all the layers that go into each feature, each component, and optimize the total solution across all that.
In our case, we run a variant of Android, we have the fork that we call Ohmni OS, and there's a tremendous amount of work just to integrate all the peripherals, like the external cameras, sensors, and even simple things like external power buttons and battery charging logic. With those considerations, you start to break out from the typical Android form factor, which is just one integrated device and you have various components that can all be connected, which becomes challenging alone. And on top of that, of course, the key challenge is having a good video streaming infrastructure.
So just like the big players, Zoom, Skype, and Teams, all these other players, we have to do a really good job of making a seamless infrastructure that is not just video and audio but also accommodates the need to drive the robot with low latency. We also need to design that video conference solution but enhance it with a very low latency control model where you can use either a gamepad, a mouse, a keyboard — any type of controller to move the robot. And as you can see, the number of things that go into this starts multiplying, it starts getting crazy.
Another thing that's really, really interesting, and difficult is audio. And, you know, we all take it for granted, people buy things like Alexa, Google Home, and we take two-way communication with devices for granted. But designing a moving platform with a really good mic, and speaker setup is a challenging task. There's a lot of acoustic design that goes into that. And the acoustic design, sometimes based on the speakers you use and the materials you use, they go back to industrial design, there are just some designs where you can't engineer a good spot for a speaker where you create enough air volume behind the speaker or isolate the speaker and microphone from each other. And all these things, add complexity.
It’s very hard to know what will work the first time you design a system like this. So, it's really important that you can prototype quickly, you're able to test in the real world, you're able to assess these kinds of things, and then evolve your product accordingly before it goes all the way to market. One of the things about us is that we're an organization that was founded on the belief that, to be a good robotics company, you need to be very lean, and you need to know how to iterate very quickly and cheaply. And that's because robotics is still in the early, early years.
Even though they’ve been around for decades it's still early and people don't just see a robot and assume, okay, this is going to solve problem X and bust out their wallets to buy one — it’s new to them, and can be scary. So, what we've done is we've created an entire team, an entire infrastructure, and a process to do robotics development 10 times faster than the traditional model. And in doing so, we're able to iterate our product extremely fast.
The 13th generation of the Ohmni robot is going to be coming out later this year, and along the way, we've learned so many lessons about what works, what doesn't work. And even today, we find surprising things out about our products and the assumptions that go into developing our products. Sometimes we assume customers want a feature, then we go test it with them, and learn they didn't care for the feature or the problem we thought we solved was a non-issue, to begin with. So, there are all these interesting learnings that you can do when you can iterate like that and put real products in front of users.
So, to summarize, the most important things we’ve found when it comes to building telepresence robots is probably the same as what most engineering companies find… it’s having a multidisciplinary team that can communicate well across all parts of the design process, engineering, production, etc. because to make a robot, all of those steps need to work and work together well. And on top of that, it is important to have a really good process for prototyping, evolving, iterating, and learning in your organization and your engineering and production process, so that you can better yourselves and better meet the needs of your target customers.
DN: What are the steps from design through production?
Phase 1: Concept and Design
This phase begins with a business discussion where we work with you to gather high-level requirements. Then we move to design idea generation where initial designs are developed as rough sketches. These sketches and presentations are shared with you, the client to review and consider for more detailed design exploration.
Following your selection of the desired idea, we further develop the high-level concepts and explore the overall contours, surfaces, color, and detail elements. These detailed sketches include feedback from the OhmniLabs production team to ensure that they are overall manufacturable and have appropriate fitment for all components.
Once you confirm the final design, we refine the design and details into 3D CAD models and provide high-quality renderings.
Phase 2: Engineering
During the engineering phase, we develop the complete system architecture from your requirements. We conduct team design reviews for both you and external teams and system-level and component trade studies. We also complete system modeling and simulation, create technical specifications and design the product development road map.
Mechanical Design & Engineering
Electrical Design & Engineering
Software Design & Engineering
CAD & analysis
Mechanical system design
Complex electro-mechanical system engineering
CAD & simulation tools
Complex custom-designed control hardware
Wiring harness schematics, design, & build
Low & high voltage power distribution design
Power electronics development
Custom Printed Circuit Board (PCB) design & assembly
Custom mobile applications
Web apps & backends
Hardware & software integration
ROS module integration
At the end of this stage, we deliver an engineering proposal, including effort and risk assessment. Together we agree on what product will be delivered, including the cost and timeline.
Phase 3: Prototyping
The goal of the prototype phase is to quickly develop a first robotic solution with which you can test out a market or a specific use case. You’ll want to consider if there is enough of a need in the market. Is the customers’ pain big enough to justify paying for your robotic solution? Is the robot usable and convenient? And is the customer willing to pay enough for your product for you to make a large enough profit?
The deliverable at this stage is a robotic solution that focuses on functions sufficient to test your MVP or minimum viable product. Below are the goals and deliverables for each of the 3 validation testing phases.
Engineering Validation Testing (EVT)
To build a product tailored to your needs and based on your specific requirements document or list based on engineering specifications with all system functions validated and integrated holistically into your design
Typically, 1-5 robots that have been custom designed and hand-built by the engineering team
Development and Validation Testing (DVT)
To test the final aesthetics of your design using rigorous systems, to get your design certified, and to finalize packaging
Typically, 10-30 custom designs and engineering-built robots which are intended to look like the final version of your product
Production Validation Testing (PVT)
To start full production using robotics technicians, contract manufacturers, vendors, and suppliers
Typically, production-ready samples followed by 1000s of market-ready robots
Phase 4: Manufacturing
At this point, you’re ready for full-speed production of your product with a full warranty, shipping, and fulfillment.
As mentioned in the initial table, the challenges here are much larger than when designing a prototype alone. At this step, we can help ensure that your product works in various environments when controlled by a range of users; that packaging is suitable for worldwide shipping; that your product receives the appropriate safety certification; and that your project is staffed with trained technicians that are continuously managed.
We can deliver the volumes at a fraction of the time and cost of traditional manufacturing, thanks to an advanced additive manufacturing process that we have been perfecting over the last five years. We build our production-grade 3D printers that have been optimized for 24/7 operation. We can spin up lean manufacturing lines while ensuring an entire quality and production process.
Phase 5: Deployment
Unlike many traditional robotics companies that only offer concept, design, and prototyping services, OhmniLabs offers end-to-end solutions that support manufacturing at scale as well as seamless deployment via Cloud robotics. This Cloud infrastructure manages all robots in the field and tracks the unit health, usage statistics, and automated upgrades with a simple click.
We help you get your robots to market fast by leveraging our comprehensive library of modular robotic tech modules/components. This collection of pre-designed, production-ready technologies include, but are not limited to:
- integrated lithium battery and power management systems
- Brushless motor control systems and direct drivetrains
- self-docking charging system
- multiple parametric chassis designs and structural components
- custom multitouch IPS HD displays and embedded systems
- long-range microphone array and acoustically tuned speaker systems
- fully integrated OS and drivers for sensors, motors, etc.
- cloud-based fleet management and update system
- deep learning and autonomy stack
- ultra-lightweight teleoperated and soon autonomous robotic arms
- certified UV-light/Disinfection kit with tech modules that are expandable on an infinite scale
Most of these components are already in mass production and we've invested the millions of dollars and thousands of person-hours required to take these components from prototype to production-ready.
On top of that, these components have been tested and refined through continual real-world usage. Every day, we're shipping and supporting these components in the production of Ohmni telepresence robots as well as in other clients' custom robots.
By leveraging our library of tech modules and lean manufacturing process, we enable you to reap the benefits of economies of scale even at lower volumes and drastically reduces risk and shortens time to market.
Depending on the scope of the customizations, however, there's often still a lot of work to be done to create a high-quality, finished product, so plan accordingly and be prepared.
DN: What type of professionals are involved in the process? What is the team like?
Jared Go: You need an army of design, engineering, production, manufacturing, inventory, cloud developers, quality assurance, quality control, product management, shipping, receiving, customs, marketing, sales, and more. Beyond that, you need support. You need success. You need fundraising. And above all, it's not just who is involved inside the company, but also externally. So, it's really important to us that the partnerships we build with our vendors, our shipping carriers, and everybody involved is an instrumental part of delivering Ohmni Robot to our customers.
You might be surprised, but when you have a robot that has hundreds of parts, inventory itself becomes such a large issue, and just the physical arrangement of that storage, tracking, everything with processes associated with that. And, you know, when you work with vendors, your vendors must deliver consistency, providing you the same part every time. Managing parts vendors is a critical job that takes an amazing amount of skill and experience. So, it's a large team and if you are ever going to make your robot, never underestimate the challenge of how many people it takes. It takes a village to make a robot.
I guess the one part that we are a little bit unique in is that our hardware team is very multidisciplinary. I think relative to a lot of traditional organizations that have very structured electrical, mechanical, maybe quality, or production teams that are a little bit more siloed, we try to have everyone cross-train and understand the challenges in each domain. And the benefit of doing that, even though it's more difficult is that each person can now make optimizations at the right level. So, I think having a unified team is unique to how we operate. And I think that is, again, a key to how we can iterate quickly and how we can build a product quickly.
DN: What are the greatest challenges in building a telepresence robot?
Jared Go: Connectivity is by far the toughest challenge when creating robots because the quality of the experience depends on it. And normally, when you have a video call or audio call, it's not terrible, you know, if the quality is a little bit laggy, it becomes inconvenient, but it's not terrible. But with robots, spotty internet could mean losing control of a moving robot in a manufacturing plant - a massive no-no! So, the bar and the standard that we hold ourselves to is much higher than a typical call. And the challenges associated with that are so many, it could simply be that the user has a weak access point. Or the way that users have deployed access points is leaves dead zones. The problem can stem from a million different places. I think anyone who's ever worked in IT will tell you, consistent Wi-Fi coverage is challenging.
Beyond connectivity, is understanding how diverse real-world conditions can be, and making sure you're strict enough about your requirements and your testing, to design well. Sometimes it takes an iteration to discover the weak points. But, to learn about and solve little issues that arise is a huge challenge you need to face head-on. You never want to learn after the fact that you’ve made thousands of defective robots. So really, also, from a process perspective, making sure you are organized, learning fast, continuously learning, assessing risk, and making sure you're finding these issues before you can scale.
DN: What are the testing procedures?
Jared Go: So, for us, the testing procedure always comes from two things. One is observing the conditions that the product will be used in, and the other is assuring operation in real-world conditions. We start by asking customers how and where they plan on using our robots. Typical questions like, How are you using it? Is it dusty? Is it bumpy? So, from that, we develop our product requirements document. And with that, every component has a functional test of its own. There are also system-wide tests that we do. And they span all different areas, of course, mechanical, electrical, even software and cloud server tests.
So, the normal testing strategy starts with us deciding on what parts are worthwhile to test and at which level. And anything more of a component test or sub-assembly test, we typically do those things on the line. We also do full system integration testing in a bunch of different ways. And we do comprehensive tests and that help us uncover defects or critical areas. It's really important for us too that the system used to track defects is part of the closed-loop design process, so that the information you get when things go wrong, immediately gets back to your other teams, and they can prioritize and work on those.
DN: What are the quality assurances involved?
Jared Go: So, we have a strict set of criteria for every robot that goes out the door. And it covers everything from fasteners and fastener torques to observed behavior. For example, wobble, displacement, fitment, screen quality, microphone, and speaker quality - basically everything that constitutes normal operation of the robot. We also regularly test our servers and web service for quality of experience because the robot itself is only part of the experience provided by a telepresence robot.
DN: If you were to summarize some key points on build a better telepresence robot, well, what would be those key points?
Jared Go: In robotics, being able to iterate fast and test in the real world is critical.
It is ideal to maintain a multidisciplinary team that can communicate well and makes design tradeoffs across all domains simultaneously, not just individually optimize each separate area, because then you'll end up with a globally sub-optimal solution.
Don't underestimate the amount of time and capital it will take to get a functional product to market. Because there's a lot of things you'll learn along the way that will force you to change the design or improve upon and so forth.
For us, we’ve found success in quick iteration, and cost-effective manufacturing through our additive manufacturing capabilities. We do 3D printing at a production scale, which allows us to create several thousands of robots a year. This allows us to do two things: one is we can change our design in a matter of hours, instead of months. If there's a defect or if some part area needs to be strengthened it gives us tremendous flexibility to go and solve those problems immediately. It also allows us to print and do certain geometry that is very challenging for typical molding processes. Leveraging all of the advantages of 3D printing is one of the things that truly make us stand out and has helped us to move faster.
Rob Spiegel has covered manufacturing for 19 years, 17 of them for Design News. Other topics he has covered include automation, supply chain technology, alternative energy, and cybersecurity. For 10 years, he was the owner and publisher of the food magazine Chile Pepper.