Robotics is an ever-evolving discipline. From manufacturing to medical robots, the applications are diverse and inspiration can come from anywhere. Take a look at five brand-new, ground-breaking innovations in robotics.
Cell-sized injectable microrobots
At only 70 micrometres long (one micrometre = one-thousandth of a millimetre), these tiny, bug-shaped robots may not be the smallest robots ever built, but that’s not the point. This size has a particular significance because it is roughly the size of human cells, according to Marc Miskin, who developed the microrobots along with colleagues at Cornell University. “In this respect, it’s not just small that counts,” Miskin explains. “We’re special because we can build robots at a fundamental scale of biology.”
The bots are also tough enough to survive in harsh environments, which opens up an exciting medical possibility: they can be injected into the body. Once there, they can manoeuvre themselves into tight spots much more easily than ordinary equipment could, and with much less risk of damaging tissue.
These microrobots don’t have an internal battery. Instead, each one is driven by four tiny solar cells, one on each leg. The legs, no more than a hundred atoms thick, are made up of dual layers of platinum and titanium. When a laser is shone on a solar cell, the platinum expands but the titanium stays rigid, causing the leg to bend. So, the robot can walk by stimulating each leg’s solar cell in turn.
The initial design of the robot is relatively simplistic, and next iterations of the design will include extra functionality, including sensors and controllers. They could also have applications outside the body. “My group is in the early stages of developing robots that live inside lithium ion batteries and clean the surface to make them safer and last longer,” says Miskin.
Grey Goo robot
The concept of a swarm of tiny robots will be familiar to many sci-fi fans, but don’t worry: the ‘particle robot’, developed by engineers at MIT and Columbia University, isn’t going to create Doctor Who’s gas mask zombies any time soon. The particle robot is a collection of ‘particles’, or units, with no centralised control system.
The particle robot, or ‘grey goo’, mimics the behaviour of human cells. Individually, cells are immobile, but as a group, they can move. How do they do this? By expanding and contracting. The cells are loosely connected, meaning that each expansion and contraction pushes and pulls on its neighbours, creating a collective movement. In the case of the particle robot, the connections are provided by weak magnets around the circumference of each unit.
In a traditional robot, the failure of a single part can render the entire machine useless. In the grey goo, the particles don’t directly communicate with or rely on any other particle, meaning that the failure of a single part has a minimal impact on the overall functionality. In fact, the researchers found that a fifth of the particles in the cluster could fail and the entire group could still travel at half-speed.
The particle robot was designed to travel towards a light source. Each particle uses its sensors to measure the intensity of light: the closer to the light source, the brighter it will appear to the particle. By broadcasting this intensity to their neighbours, the particles can co-ordinate their expansions and work together to wriggle towards their goal.
The flexibility of the robot is its real strength. Since it is made up of a group of universal cells, the collective can adapt as a whole to tackle a variety of tasks. This means that the particles can change their shape to fit through an obstacle or surround an object to transport it.
Robots are fantastic at performing identical, repetitive tasks. When it comes to tasks like picking up lots of identical products and placing them into packaging, robots are ideal. The trouble comes in when you want a robot that can pick up different types of object. Unlike human hands, traditional robot grippers need to be designed precisely for the objects they pick up to make sure they grasp them securely.
Roboticists at Harvard University and MIT, led by Shuguang Li, have developed an alternative: a soft, origami-inspired gripper that can lift a huge variety of objects. Its silicone rubber skeleton is covered with a latex skin. The bell-shaped hand lowers onto an object, and the air is sucked out of the skin, pulling the skeleton down into its folded origami shape.
The gripper delicately takes hold of objects, relying on the friction of the ruched skin to hold them in place. Since it lacks the rigidity of a traditional robot gripper, it can hold fragile objects like soft fruit or glass without damaging them. The gripper can also hold objects of a wide variety of shapes and materials, including a rubber duck, a gaming controller and a watering can.
Backflipping Mini Cheetah
In 2017, we saw Boston Dynamics’ humanoid robot Atlas backflipping off a box, stick the landing, and raise its arms in triumph. Not to be outdone, MIT has developed a mini version of its Cheetah robot, and this tiny quadruped can backflip too. In fact, Mini Cheetah is the first four-legged robot to perform a backflip.
Backflipping might not be the most useful skill for a robot to have, but it does display Mini Cheetah’s extraordinary ability to react to its environment. It can land on its feet when dropped, recover from a knock without falling over, and right itself when it does end up the wrong way up. It can also display a range of movements, including a forwards-facing trot-run, spinning whilst moving, and an antelope-like bounce called ‘pronking’.
Dactyl robotic hand
OpenAI, San Francisco-based AI research organisation, has approached the task of a robot gripper in an entirely different way. This gripper, named Dactyl, is based on the human hand, coupled with a system of cameras. Dactyl uses its visual input from the cameras to assess the object in its hand.
Dactyl uses neural networks to learn how to manipulate the object in its hand: it is ‘trained’ with huge amounts of data until it can recognise patterns well enough to predict what comes next. Through this technique, it has learned how to turn a toy block in its hand until it matches a certain orientation.
The tricky part with training robots is the vast difference between simulation and reality. Simulations are nice and tidy, but reality can be messy: for one thing, friction makes the situation much more complex. To make up for this, OpenAI intentionally introduced a level of randomness into the data they provided for Dactyl to add a touch of realism.