Humanoid Robot Safety Comes Into Focus
This article provides an in-depth look at topics related to Autonomous Systems.
For expert-level training—including certificate-based programs—on these topics and more, explore our training programs. To learn how we support product development, compliance, and organizational safety goals with consulting support, visit our consulting page—or contact us directly.
We know they’re coming and we know they need safety. How will we achieve it?
The humanoid robots are nearly here. You can find them running, cleaning up, folding clothes and putting away the dishes. I guess some robots are boxing enthusiasts also. The boxing seems a bit dangerous to a safety engineer. After all, some of these mechanical fellows weigh in at over 200 pounds. And unlike real humans who are a bit soft and cushiony, humanoid robots tend to be made of the hard stuff. You wouldn’t want that metal foot stepping on your toe, much less the whole weight of a robot falling on top of you.
The Need for New Safety Paradigms
For safety purposes, we can keep ourselves out of the robot boxing ring. In fact in a lot of traditional industrial settings, safety strategy is built around exactly this idea. Just keep humans out of the robot rooms, so that humans can’t be hurt by robot motions, accidents, falls, malfunctions etc. We can gain a lot of safety with little complexity, using this time-tested idea. For traditional stationary robots in industrial environments, we use light curtains to detect any human entering the robotic cell. If a human enters, the light curtain is interrupted, and the robot shuts down immediately. Problem solved.
But this kind of safety strategy won’t be tenable for most humanoid robots. After all, the point of humanoid robots is to operate with humans. Be it in the home or the workplace, humanoid robots are intended to assist us inside our own human environments. To do this, the robots need to share our space. They can’t be separated by walls or light-curtains. Instead they’ll co-exist with us, working together, literally hand-in-hand. That’s the vision and it’s rapidly becoming reality. How can we make it safe?
To safety-proof the robot-enabled future, we’ll need some robot-ready safety tools, methods, and ideas. The good news is, some of those ideas are already well-honed from other industries where autonomous technology has gained real-world traction. In autonomous drones, self-driving vehicles, and mobile industrial robots, we’re already deploying good safety ideas to make applications safe. And robots can adapt those ideas to enable safe humanoids who interact with humans, seamlessly and safely.
Three Essential Safety Ideas for Coexistence
Here’s a starter pack of three big safety ideas that we can use to ensure safe humanoid robots:
1. Controlled shut-down is a must: In many traditional simple systems, the so-called “safe state” is simply a shut-down. An example of this is the well-known Emergency Stop Button…. You press it; the system powers down ASAP. Simple, right? But that won’t be quite enough for a humanoid robot. What if it’s moving and you shut it down mid-step? Humanoid robots are actively balancing themselves; just as you would fall if you “froze” mid-step, so too will a humanoid robot fall down (and maybe injure bystanders) if it goes into immediate shutdown mode. That can’t be allowed to happen.
To address this safety requirement, robotic motion controls will need to to achieve a safe and balanced position when shutting down fast. That requires some level of redundancy… not full-blown 100% redundant-everything, but a limited set of functions that can keep balance even after faults, failures, or shut-down commands. This version of a limited “fail-operational” concept is already in force in autonomous functions where sudden shut-down isn’t an option (think self-driving cars for example… in case of a malfunction it’s not an option to just “switch off” the autonomous driver as it rolls down the road at 50mph). Notably, the control functions that keep a robot stable may only require limited motions and short timeframes of fail-operational capability, all of which helps keep costs down. There’s a lot of work to do, and we expect to see diverse solutions to the problem. But in the end, controlled shut-down with fail-operational systems will soon become a ticket-to-entry for interactive humanoid robotics.
2. V & V means scenarios, scenarios, scenarios: One thing we’ve learned from autonomous systems engineering is that safety requires safety test and confirmation… formally known as verification and validation (V&V). And the scale of that safety V&V activity can be mind-bending. Imagine thousands or millions of scenario variants, with the robot interacting with humans in all kinds of scenarios with potential safety implications. Robot walking through a room with small children. Robot pouring hot tea. Robot putting knives away. And so on and so forth; all with zillions of variations in environmental parameters such as lighting, room geometry, motion of the robot, motion of other humans, motions of the family dog…. the list goes on and on.
To really prove a robot is safe, we need to expose it to all these scenarios and many more. To do this, two suggestions are useful. First: a digital twin or simulation-space must be used to get a wide sweep of scenario V&V. Although testing is necessary, we simply can’t test every plausible scenario… we need to simulate. It’s not a choice between testing and simulation. Both are required. The second key idea is to build an organizational framework to manage all these scenarios. That’s a non-trivial challenge, as different enterprises may break down scenario organizations and ontologies in diverse ways. The smart robotics players will be figuring that out sooner rather than later.
3. An international standard is must-have: Safety standards for autonomous systems are among the most misunderstood documents of our time. Most take a risk-based approach to safety. What does that mean? It means that instead of prescribing exact technical measurables or tests to be achieved, the standard instead provides a framework by which any autonomous system (drone, car, humanoid-robot, etc.) can be safely developed based on best practices, safety-related processes, and safety architectural ideas. You don’t perform one test to show compliance to a risk-based standard. You show adherence to processes and metrics through the development lifecycle.
We’ll see this in the forthcoming ISO-25785-1 standard, now in development for robots including mobile manipulator robots with actively controlled stability. (That’s technical jargon for our humanoid dishwashers, mentioned above). This standard should serve as a global benchmark that brings humanoid robotic safety under a sensible internationally agreed standard. That will be far better than cobbling together ideas from the many existing functional safety standards in use (such as IEC61508, ISO 13842, ISO13849, ISO26262, ISO 21448, and many others). The smart players will learn to use this new ISO-25785-1 standard when it arrives, understanding the framework in depth and tailoring requirements to their unique humanoid robotic use cases.
A Standardized Future
After all, just as each human is unique, each model of humanoid robot will bring different capabilities to the table. Lets bring all that capability to market safely, in accordance with ISO-25875-1, and with strong scenario V&V and fail-operational balance capabilities. That’s how we’ll know that the robot’s pugilist tendencies are confined to the boxing ring where they belong.
Have insights or questions? Send us an email at info@sres.ai or leave a comment below—we welcome thoughtful discussion from our technical community.
Interested in learning more about our services? Find all of our upcoming training programs here and all of our consulting offerings here.
