In the world of robotics, making fast, smart decisions is crucial.
A new algorithm developed by researchers at Caltech is helping robots plan their movements in real-time, no matter what type of robot they are.
This system, called Spectral Expansion Tree Search (SETS), allows robots to quickly choose the best actions to take while navigating their environments.
The idea for SETS comes from a game-playing algorithm used by Google DeepMind’s AlphaZero, which taught itself to master chess, shogi, and Go.
Like AlphaZero, SETS works by exploring different possibilities and selecting the best one. But while AlphaZero’s focus is on winning board games, SETS is designed for robots moving in the real world.
How does SETS work?
Robots, like drones or self-driving cars, often face complex challenges when moving through their surroundings.
They need to avoid obstacles, adapt to unexpected events, and make decisions quickly.
Traditionally, these movements were carefully programmed by designers. However, with SETS, robots can figure out the best path themselves.
SETS uses a concept called a Monte Carlo Tree Search. Think of it like a decision tree, where the robot looks at different possible movements and their outcomes.
However, simulating every possible movement in the real world would take years. This is where SETS stands out: it balances two strategies—exploration and exploitation.
- Exploration: Trying out new paths or movements to see if they might work better.
- Exploitation: Focusing on paths that have already shown good results.
By combining these strategies, SETS quickly zeroes in on the best solution without wasting time on bad options, like paths that lead to crashing into a wall. The algorithm runs this process thousands of times in just a fraction of a second, allowing robots to make decisions continuously.
To show how versatile SETS is, the researchers tested it on three very different robotic systems:
- A Drone: The drone navigated an airfield filled with turbulence and avoided obstacles while tracking certain objects. This experiment, conducted at Caltech’s Center for Autonomous Systems and Technologies (CAST), showed how SETS helps robots adapt to unpredictable conditions.
- A Ground Vehicle: The algorithm assisted a tracked vehicle, similar to a tank, in safely navigating a narrow, winding track without hitting the sides.
- Tethered Spacecraft: SETS helped two connected spacecraft capture and redirect a third object, mimicking scenarios like rescuing a damaged satellite or redirecting an asteroid.
Each test demonstrated how SETS can handle unique challenges without needing any special adjustments.
What makes SETS special?
Unlike traditional approaches, SETS doesn’t require programmers to customize the system for each robot.
The algorithm can be applied to almost any type of robotic platform, making it a game-changer in the field of robotics.
Soon-Jo Chung, a senior researcher at Caltech and NASA’s Jet Propulsion Laboratory, highlights the innovation: “Our algorithm strategizes, explores all possible motions, and selects the best one quickly, something other methods can’t do efficiently.”
The team is already working on applying SETS to a race car competing in the Indy Autonomous Challenge at the Consumer Electronics Show (CES) in January.
With SETS, robots are not only getting smarter but also faster, more adaptable, and ready to tackle real-world challenges. This breakthrough could lead to safer drones, better self-driving cars, and even smarter robots in homes and space.
