Imagine the scene after a huge earthquake or natural catastrophe, of the type we’ve seen in Fukushima or Haiti, an injured victim or terrified child pinned under rubble. After some jostling a spotlight pierces the darkness, the sound of hydraulics and motors approaches, and the rubble is lifted safely clear by a rescuer who isn’t even human…
Advances in robots are making many expert predict a (near) future where search and rescue robots will penetrate disaster zones en masse. But their success depends on the alignment of several disciplines.
Going all day
The first thing a robot plunging into danger needs to do is power itself independently, and for as long as necessary. It takes a lot of electrical power to drive devices that are usually so heavy, and the heavier they are, the more power they have to carry on board (which makes them heavier, which requires more power in turn, and so on…)
Robot energy solutions vary greatly. Boston Dynamics’ BigDog carries a one-cylinder, two stroke Go-Kart engine (think of a lawnmower), which drives 16 hydraulic motors in its legs. The Mars Curiosity Rover can theoretically keep going forever (provided the mechanisms still work) as it recharges itself with a solar panel.
In the world’s foremost robotics competitions, entrants can’t be tethered to external power or communications, and in the toughest challenges wireless communication is purposefully degraded to give them a chance to prove their self-help skills.
While that seems tough, a city struck by a killer earthquake or a forest engulfed in flames will be even tougher. Search and recue bots will have to go deep into dangerous territory, cut off from human operators with patchy communication signals and having to carry their own power sources.
The other challenge for a robot untethered from operators’ instructions will be making its own decisions about the next course of action, using machine learning and other AI algorithms to self-teach. Pre-programming robots for unpredictable environments is virtually impossible, but the alternative can be just as dangerous. There’s a sweet spot to be found, and ‘learning to unlearn’ certain behaviours can be just as important in the field.
Restrict self-learning too much and the simplest thing might become a fatal stumbling block, like a flight of stairs or a door handle. Trust a system too much to try new things and it might decide a disaster victim is another piece of rubble and cause more harm.
The other secret to successful search and rescue is sensors, and there are as many kinds as there are environments they have to work in. Many designers consider a motion tracker essential, the kind we find in video game controllers or smartphones. With feedback from accelerometers or gyroscopes in multiple dimensions, motion sensors give the robot critical information like orientation to the ground – a big deal when scrambling over wreckage.
In fact, some of the most basic sensors (like the motion tracker) inform on movement. If you’re talking about a humanoid shaped robot, load-bearing sensors have to measure the shifts in weight and direct actuators (motors) to compensate to move the body in the other direction to keep it upright.
For robots that are connected to operators at a home base, visual sensors are also crucial. Cameras – often two of them to provide an appreciation of depth, like our eyes do – can show the operator what’s going on in the immediate area.
We can also design robots with sensors for dangers they’re likely to come across in specific environments. The Sandia Laboratories’ Gemini-Scout is designed for mining accidents, finding and delivering provisions to survivors. As well as the ability to scramble over rocks, debris, tracks – even through water and mud – it has a thermal imager to acquire video, a speaker and microphone for communication and temperature and gas sensors so it knows what it’s getting into (if the worst happens, it’s also explosion proof).
Terrain is among the biggest challenges search and rescue robots face. Where most robots that move are designed for the flat, even ground in a lab, the surfaces in the real world – even the footpath outside your house – demand constant unconscious shifts in balance and weight to navigate, one the human nervous system is uniquely suited to.
After the destruction of a major disaster it can be even harder, so the challenge is often to design the best method of locomotion. Wheels are of limited use (although unique configurations of movable wheel arrays are catching on – see ‘The Fukushima Nuclear Disaster’). Designs inspired by quadruped animals like Boston Dynamics’ BigDog are also showing promise.
Humanoid robots seem the natural choice to scramble into wreckage, but a huge amount of processing and actuator power is devoted to just keeping it standing (as it is in us, using the brain and muscles), which takes power away from the task at hand.
Disaster robots got their first real debut when they were sent into the incredibly difficult terrain of the World Trade Centre towers following the September 11 attacks. They didn’t perform at all well, often getting stuck or breaking, but the experience gave engineers a lot of real-world experience to work on the next generation of rescue bots.
But here’s another wrinkle. After spending all that development time and money on a single machine (to say nothing of the danger and heartbreak it’s trying to save victims from) only to have it crushed flat by a falling wall or run out of power at the worst possible moment and be lost forever, the answer might be to not put all your eggs in one basket.
The solution for some environments might be a hive of redundant units – a robot swarm. If there’s a lot of thick concrete or metal at the disaster site, communication is likely to be very unreliable, so if a comms link is lost with one individual bot, it can be maintained along a chain between those that are still in range, the command passed down the line to the unit at the coal face.
A swarm also allows for a distributed processing model. Each unit has their piece of the puzzle but is aware of the outlook of every other bot and can take over the decision-making or operator-response should something happen to its nearby fellows. It’s a little like having one giant robot body and brain made up of small, fluid elements. In some cases the robots themselves can create the communications network for human responders – or other robots – on the ground, like the Smavnet project of inexpensive, autonomous Micro Air Vehicles (MAVs).
Robot swarms also allow for the opportunity to build and deploy several kinds of bot, each with its own talents. Larger, longer-range robots could carry smaller and more specialised devices like the Snakebot deep into a disaster zone to go to work.
One snakebot model by Japanese robotics professor Satoshi Tadokoro is 26 feet long and propels itself using nylon bristles powered by tiny individual motors. It only moves at a crawl of two inches per second, but it can climb 20 degree inclines, turn very sharp corners and see what’s ahead with its front-mounted camera.
Even though low-powered snake-inspired robots are built for localised environments, there’s no reason why bigger, longer-range models couldn’t carry them to the burned out factory or collapsed building and deploy them to map and report back on the environment.
Whatever the shape or size, search and rescue robots might accompany human rescuers into dangerous conditions or they might comprise the first wave of responders alone, but few have any doubt they’re right around the corner.