When the robots of the future are set to extract minerals from other planets, they need to be both self-learning and self-repairing. Researchers at Oslo University have already succeeded in producing self-instructing robots on 3D printers.
On the third floor of the Department of Informatics there is a robotics laboratory which looks like a playroom This is where researchers are testing how their robots can figure out how to move past barriers and other obstacles.
“In the future, robots must be able to solve tasks in deep mines on distant planets, in radioactive disaster areas, in hazardous landslip areas and on the sea bed beneath the Antarctic. These environments are so extreme that no human being can cope. Everything needs to be automatically controlled. Imagine that the robot is entering the wreckage of a nuclear power plant. It finds a staircase that no-one has thought of. The robot takes a picture. The picture is analysed. The arms of one of the robots is fitted with a printer. This produces a new robot, or a new part for the existing robot, which enables it to negotiate the stairs,” hopes Associate ProfessorKyrre Glette who is part of the Robotics and intelligent systems research team at Oslo University’s Department of Informatics.
Three generations
Even if Glette’s ideas remain visions of the future, the robotics team in the Informatics Building have already developed three generations of self-learning robots.
Professor Mats Høvin was the man behind the first model, the chicken-robot named “Henriette”, which received much media coverage when it was launched ten years ago.
Henriette had to teach itself how to walk, and to jump over obstacles. And if it lost a leg, it had to learn, unaided, how to hop on the other leg.
A few years later, Masters student Tønnes Nygaard launched the second generation robot. At the same time, the Informatics team developed a simulation program that was able to calculate what the body should look like. Just as for Henriette, its number of legs was pre-determined, but the computer program was at liberty to design the length of the legs and the distance between them.
The third generation of robots brings even greater flexibility. The simulation programme takes care of the complete design and suggests the optimal number of legs and joints.
“We tell the simulation program what we would like the robot to do, how fast it should walk, its size and energy consumption. For instance, we may want the robot to be able to turn around and change direction, climb over boulders and walk on rugged ground,” says Kyrre Glette.
In response, the simulation program suggests the optimal solution, including the shape of the body and the number of legs. It simulates thousands of possibilities and produces the best models by means of artificial evolution.
In other words, the computer program comes up with a set of proposals for different robots designed for optimal functionality.
“None of them is intrinsically better than the others. The only difference is their strategies for solving the tasks,” says research fellow Eivind Samuelsen.
Unfortunately, it is difficult to arrive at the perfect solution by way of simulation, because our wish-list of what we want the robots to do is becoming increasingly complicated.
Robot trials
Simulation is not enough. In order to test the functionality of the robots, they need to undergo trials in the real world. The robots are produced as printouts from a 3D printer.
“Once the robots have been printed, their real-world functionalities quite often prove to be different from those of the simulated versions. We are talking of a reality gap. There will always be differences. Perhaps the floor is more slippery in reality, meaning that the friction coefficient will have to be changed. We are therefore studying how the robots deteriorate from simulation to laboratory stage,” says Mats Høvin.
The robots are frequently only half as good in real life.
“Practice enables the robots’ performance to be improved by 20 to 40 per cent. We hope that gradually, we will be able to double their performance ratings so that they end up being just as good as, or perhaps even better than their simulated counterparts.”
The researchers are looking at the robots’ ability to learn and practice even at the simulation program stage.
“We then proceed to check whether the robots are able to learn in a real-world setting. One of the tests is to see how the robots react to obstacles,” says Samuelsen.
From three to six legs
The robotics team are currently comparing the performance of five robots which in theory should be equally good.
Three of the five robots have four legs, one has three, another has six. The fewer legs, the less energy is consumed. One of the robots is fitted with single-joint legs. The others have legs with two joints.
“It may well be that the best solution would be to use three joints, but this would mean that the energy consumption would be far too high.”
Just like in the natural world, the researchers have therefore opted to restrict the number of joints to two.
“Single-jointed legs enable the robot to walk forward efficiently, but it will find it difficult to turn around,” Glette points out.
When the researchers are testing the robots, they set up an obstacle course to enable the robots to teach themselves how to pass the hurdles.
The researchers hope that in the future, the robots will be able to give automatic feedback to the simulation program about how well they work, so that the next time around, the computer will be in a position to design an even better robot.
“Once we have further perfected the robots, we will also be testing how they work if printed in a softer material.”
3D printers
3D printers make it possible to create sophisticated robots and other objects.
The Latest on: Self-learning robots
[google_news title=”” keyword=”Self-learning robots” num_posts=”10″ blurb_length=”0″ show_thumb=”left”]
via Google News
The Latest on: Self-learning robots
- A better way to control shape-shifting soft robotson May 10, 2024 at 10:52 am
A new machine-learning technique can train and control a reconfigurable soft robot that can dynamically change its shape to complete a task. The researchers also built a simulator that can evaluate ...
- 6 top-rated robot vacuums of 2024on May 10, 2024 at 8:30 am
Automate your cleaning routine with these staff and expert recommended robot vacuums from brands like Roomba, Roborock, Shark and more.
- Dexterous robot hand can take a beating in the name of AI researchon May 10, 2024 at 6:10 am
A robotics company likely most famous for a demo of its dexterous robot hand at ICRA 2019 with Jeff Bezos has now unveiled a new robust model designed for machine learning research, which was ...
- The week's best robot vacuum deals: Roomba and Eufyon May 9, 2024 at 10:00 am
Learn more about how we select deals ... The chic gray color scheme is also unique in a world of black-and-white robot vacuums. Roborock Q5+ with self-empty station — $359.99 $699.99 (save ...
- ‘Indestructible’ 3-fingered robot hand being tested by Google’s Deepmindon May 9, 2024 at 7:16 am
Reports suggest that Googe DeepMind is using a new dexterous 3-fingered robotic hand from Shadow Robot to train AI systems for robotics.
- Exclusive: Wayve co-founder Alex Kendall on the autonomous future for cars and robotson May 9, 2024 at 6:59 am
U.K.-based autonomous vehicle startup Wayve started life as a software platform loaded into a tiny electric "car" called Renault Twizy. Festooned with cameras, the company's co-founders and PhD ...
- Wayve raises $1B to take its Tesla-like technology for self-driving to many carmakerson May 9, 2024 at 6:45 am
Wayve, a U.K.-born startup developing a self-learning rather than rule-based system for autonomous driving, has closed $1.05 billion in Series C funding ...
- The cheapest robot vacuum sales and deals for May 2024on May 7, 2024 at 5:00 pm
The Roomba also features smart mapping so the vacuum will learn what rooms ... i7's more expensive mapping and self-cleaning features. There are also some excellent robot vacuum sales on the ...
- ChatGPT trains robot dog to walk on Swiss ballon May 6, 2024 at 7:50 pm
This quadruped wobbling along balanced on top of a yoga ball is a fun experiment to watch – but at its core, it demonstrates that AIs like GPT-4 can train robots to perform complex, real-world tasks ...
- MaxDiff RL Algorithm Improves Robotic Learning with “Designed Randomness”on May 6, 2024 at 1:58 pm
In a groundbreaking development, engineers at Northwestern University have created a new AI algorithm that promises to transform the field of smart robotics. The algorithm, named Maximum Diffusion ...
via Bing News