The AI controls a vaguely humanoid-like creature, albeit with a heavily-simplified body and limbs. It “lives” in a 3D environment created in the Unity engine, which provides the necessary physics engine for the work. Meanwhile, the ML-Agents package is used to provide the brain for Albert, the AI charged with learning to walk.
The video steps through a variety of “deep reinforcement learning” tasks. In these, the AI is rewarded for completing goals which are designed to teach it how to walk. Albert is given control of his limbs, and simply charged with reaching a button some distance away on the floor. After many trials, he learns to do the worm, and achieves his goal.
Getting Albert to walk upright took altogether more training. Lumpy ground and walls in between him and his goal were used to up the challenge, as well as encouragements to alternate his use of each foot and to maintain an upright attitude. Over time, he was able to progress through skipping and to something approximating a proper walk cycle.
One may argue that the teaching method required a lot of specific guidance, but it’s still a neat feat to achieve nonetheless. It’s altogether more complex than learning to play Trackmania, we’d say, and that was impressive enough in itself. Video after the break.
If [jfarro]’s project looks like a miniature version of the much-touted solar freakin’ roadways concept, rest assured that there are huge differences. For one, these lighted pavers actually work — trust me on this; I live not far from the demo site for the Solar Roadways and the degree to which it underwhelms cannot be overstated. Granted, a garden path is a lot simpler to engineer than a road, but many of the challenges remain.
Using recycled glass blocks that are usually reserved for walls and windows, [jfarro] figured out how to attach Neopixel rings to the underside and waterproof them with a silicone conformal coating. The 12 lighted pavers he built draw considerable current, so a 45-watt solar array with charge controller and battery were installed to power the pavers. An Arduino and a motion sensor control the light show when someone approaches; more complicated programs are planned.
Hats off the [jfarro] for taking on a project like this. We don’t often see builds where electrical engineering meets civil engineering, and even on a small scale, dealing with dirt, stone, and water presents quite a few challenges. Here’s hoping his project lasts longer than the Solar Roadways project did.
[Becky Stern] came up with a way to make sure you and your dog are getting enough exercise. It’s a dog collar mounted GPS that measures how far you have walked. Just set your target distance and the progress bar in the middle of this flower will let you know when you reached it.
The most obvious piece of hardware is the OLED board which is sticking out like a sore thumb. But if you’d like to be a little more discreet you could forego the full-featured display for some carefully places LEDs to make up a circular progress bar. The GPS module itself fits well in the center of the flower, which [Becky] shows us how to make out of wire-edged ribbon. Hidden on the other side is an ATmega23u4 breakout board running the Arduino bootloader.
He was inspired to send in a link to his research after reading about the Kinect-powered shopping card robot. He figures that that project is similar to his own, but his does have several added benefits. The first being that if a robot is on a leash, everyone knows who that bot is following or assisting. But there is the added benefit of the user needing no training whatsoever. That’s because the act of walking a dog on a leash is commonplace in developed societies; you may not have ever owned a dog, but you’ve seen others walking them on leashes numerous times and could do so yourself without any training.
The leash connects to a sensor-filled turret in the center of the robot’s body. The bot can sense when, and in which direction the user is pulling the leash. There’s also an emergency kill switch on the handle for added functionality. Take a look at some of the test video after the break to see how quickly humans can adapt to this type of user interface.
[Robert Lam] has produced a number of video tutorials, his latest being a tutorial on how to make a biped robot walk. He is mainly covering the individual motions and actions. He doesn’t go into any specific programming, but rather breaks down the act of walking into several motions and discusses the reason you need them as well as some variations. For some this will seem like obvious observations, but we’ve seen plenty of biped robots that attempt to walk without shifting their weight.You can watch this video after the break, but be sure to dig around in some of his other tutorials for plenty of good stuff.