Tag Archives: action
#439032 To Learn To Deal With Uncertainty, This ...
AI is endowing robots, autonomous vehicles and countless of other forms of tech with new abilities and levels of self-sufficiency. Yet these models faithfully “make decisions” based on whatever data is fed into them, which could have dangerous consequences. For instance, if an autonomous car is driving down a highway and the sensor picks up a confusing signal (e.g., a paint smudge that is incorrectly interpreted as a lane marking), this could cause the car to swerve into another lane unnecessarily.
But in the ever-evolving world of AI, researchers are developing new ways to address challenges like this. One group of researchers has devised a new algorithm that allows the AI model to account for uncertain data, which they describe in a study published February 15 in IEEE Transactions on Neural Networks and Learning Systems.
“While we would like robots to work seamlessly in the real world, the real world is full of uncertainty,” says Michael Everett, a post-doctoral associate at MIT who helped develop the new approach. “It's important for a system to be aware of what it knows and what it is unsure about, which has been a major challenge for modern AI.”
His team focused on a type of AI called reinforcement learning (RL), whereby the model tries to learn the “value” of taking each action in a given scenario through trial-and-error. They developed a secondary algorithm, called Certified Adversarial Robustness for deep RL (CARRL), that can be built on top of an existing RL model.
“Our key innovation is that rather than blindly trusting the measurements, as is done today [by AI models], our algorithm CARRL thinks through all possible measurements that could have been made, and makes a decision that considers the worst-case outcome,” explains Everett.
In their study, the researchers tested CARRL across several different tasks, including collision avoidance simulations and Atari pong. For younger readers who may not be familiar with it, Atari pong is a classic computer game whereby an electronic paddle is used to direct a ping pong on the screen. In the test scenario, CARRL helped move the paddle slightly higher or lower to compensate for the possibility that the ball could approach at a slightly different point than what the input data indicated. All the while, CARRL would try to ensure that the ball would make contact with at least some part of paddle.
Gif: MIT Aerospace Controls Laboratory
In a perfect world, the information that an AI model is fed would be accurate all the time and AI model will perform well (left). But in some cases, the AI may be given inaccurate data, causing it to miss its targets (middle). The new algorithm CARRL helps AIs account for uncertainty in its data inputs, yielding a better performance when relying on poor data (right).
Across all test scenarios, the RL model was better at compensating for potential inaccurate or “noisy” data with CARRL, than without CARRL.
But the results also show that, like with humans, too much self-doubt and uncertainty can be unhelpful. In the collision avoidance scenario, for example, indulging in too much uncertainty caused the main moving object in the simulation to avoid both the obstacle and its goal. “There is definitely a limit to how ‘skeptical’ the algorithm can be without becoming overly conservative,” Everett says.
This research was funded by Ford Motor Company, but Everett notes that it could be applicable under many other commercial applications requiring safety-aware AI, including aerospace, healthcare, or manufacturing domains.
“This work is a step toward my vision of creating ‘certifiable learning machines’—systems that can discover how to explore and perform in the real world on their own, while still having safety and robustness guarantees,” says Everett. “We'd like to bring CARRL into robotic hardware while continuing to explore the theoretical challenges at the interface of robotics and AI.” Continue reading
Posted in Human Robots
#438798 This AI Thrashes the Hardest Atari Games ...
Learning from rewards seems like the simplest thing. I make coffee, I sip coffee, I’m happy. My brain registers “brewing coffee” as an action that leads to a reward.
That’s the guiding insight behind deep reinforcement learning, a family of algorithms that famously smashed most of Atari’s gaming catalog and triumphed over humans in strategy games like Go. Here, an AI “agent” explores the game, trying out different actions and registering ones that let it win.
Except it’s not that simple. “Brewing coffee” isn’t one action; it’s a series of actions spanning several minutes, where you’re only rewarded at the very end. By just tasting the final product, how do you learn to fine-tune grind coarseness, water to coffee ratio, brewing temperature, and a gazillion other factors that result in the reward—tasty, perk-me-up coffee?
That’s the problem with “sparse rewards,” which are ironically very abundant in our messy, complex world. We don’t immediately get feedback from our actions—no video-game-style dings or points for just grinding coffee beans—yet somehow we’re able to learn and perform an entire sequence of arm and hand movements while half-asleep.
This week, researchers from UberAI and OpenAI teamed up to bestow this talent on AI.
The trick is to encourage AI agents to “return” to a previous step, one that’s promising for a winning solution. The agent then keeps a record of that state, reloads it, and branches out again to intentionally explore other solutions that may have been left behind on the first go-around. Video gamers are likely familiar with this idea: live, die, reload a saved point, try something else, repeat for a perfect run-through.
The new family of algorithms, appropriately dubbed “Go-Explore,” smashed notoriously difficult Atari games like Montezuma’s Revenge that were previously unsolvable by its AI predecessors, while trouncing human performance along the way.
It’s not just games and digital fun. In a computer simulation of a robotic arm, the team found that installing Go-Explore as its “brain” allowed it to solve a challenging series of actions when given very sparse rewards. Because the overarching idea is so simple, the authors say, it can be adapted and expanded to other real-world problems, such as drug design or language learning.
Growing Pains
How do you reward an algorithm?
Rewards are very hard to craft, the authors say. Take the problem of asking a robot to go to a fridge. A sparse reward will only give the robot “happy points” if it reaches its destination, which is similar to asking a baby, with no concept of space and danger, to crawl through a potential minefield of toys and other obstacles towards a fridge.
“In practice, reinforcement learning works very well, if you have very rich feedback, if you can tell, ‘hey, this move is good, that move is bad, this move is good, that move is bad,’” said study author Joost Huinzinga. However, in situations that offer very little feedback, “rewards can intentionally lead to a dead end. Randomly exploring the space just doesn’t cut it.”
The other extreme is providing denser rewards. In the same robot-to-fridge example, you could frequently reward the bot as it goes along its journey, essentially helping “map out” the exact recipe to success. But that’s troubling as well. Over-holding an AI’s hand could result in an extremely rigid robot that ignores new additions to its path—a pet, for example—leading to dangerous situations. It’s a deceptive AI solution that seems effective in a simple environment, but crashes in the real world.
What we need are AI agents that can tackle both problems, the team said.
Intelligent Exploration
The key is to return to the past.
For AI, motivation usually comes from “exploring new or unusual situations,” said Huizinga. It’s efficient, but comes with significant downsides. For one, the AI agent could prematurely stop going back to promising areas because it thinks it had already found a good solution. For another, it could simply forget a previous decision point because of the mechanics of how it probes the next step in a problem.
For a complex task, the end result is an AI that randomly stumbles around towards a solution while ignoring potentially better ones.
“Detaching from a place that was previously visited after collecting a reward doesn’t work in difficult games, because you might leave out important clues,” Huinzinga explained.
Go-Explore solves these problems with a simple principle: first return, then explore. In essence, the algorithm saves different approaches it previously tried and loads promising save points—once more likely to lead to victory—to explore further.
Digging a bit deeper, the AI stores screen caps from a game. It then analyzes saved points and groups images that look alike as a potential promising “save point” to return to. Rinse and repeat. The AI tries to maximize its final score in the game, and updates its save points when it achieves a new record score. Because Atari doesn’t usually allow people to revisit any random point, the team used an emulator, which is a kind of software that mimics the Atari system but with custom abilities such as saving and reloading at any time.
The trick worked like magic. When pitted against 55 Atari games in the OpenAI gym, now commonly used to benchmark reinforcement learning algorithms, Go-Explore knocked out state-of-the-art AI competitors over 85 percent of the time.
It also crushed games previously unbeatable by AI. Montezuma’s Revenge, for example, requires you to move Pedro, the blocky protagonist, through a labyrinth of underground temples while evading obstacles such as traps and enemies and gathering jewels. One bad jump could derail the path to the next level. It’s a perfect example of sparse rewards: you need a series of good actions to get to the reward—advancing onward.
Go-Explore didn’t just beat all levels of the game, a first for AI. It also scored higher than any previous record for reinforcement learning algorithms at lower levels while toppling the human world record.
Outside a gaming environment, Go-Explore was also able to boost the performance of a simulated robot arm. While it’s easy for humans to follow high-level guidance like “put the cup on this shelf in a cupboard,” robots often need explicit training—from grasping the cup to recognizing a cupboard, moving towards it while avoiding obstacles, and learning motions to not smash the cup when putting it down.
Here, similar to the real world, the digital robot arm was only rewarded when it placed the cup onto the correct shelf, out of four possible shelves. When pitted against another algorithm, Go-Explore quickly figured out the movements needed to place the cup, while its competitor struggled with even reliably picking the cup up.
Combining Forces
By itself, the “first return, then explore” idea behind Go-Explore is already powerful. The team thinks it can do even better.
One idea is to change the mechanics of save points. Rather than reloading saved states through the emulator, it’s possible to train a neural network to do the same, without needing to relaunch a saved state. It’s a potential way to make the AI even smarter, the team said, because it can “learn” to overcome one obstacle once, instead of solving the same problem again and again. The downside? It’s much more computationally intensive.
Another idea is to combine Go-Explore with an alternative form of learning, called “imitation learning.” Here, an AI observes human behavior and mimics it through a series of actions. Combined with Go-Explore, said study author Adrien Ecoffet, this could make more robust robots capable of handling all the complexity and messiness in the real world.
To the team, the implications go far beyond Go-Explore. The concept of “first return, then explore” seems to be especially powerful, suggesting “it may be a fundamental feature of learning in general.” The team said, “Harnessing these insights…may be essential…to create generally intelligent agents.”
Image Credit: Adrien Ecoffet, Joost Huizinga, Joel Lehman, Kenneth O. Stanley, and Jeff Clune Continue reading
Posted in Human Robots