# Tag Info

23

It's certainly possible to implement this on an Arduino. Here are 3 such Arduino libraries that implement neural networks: Neuroduino Arduino Basics ArduinoANN The complexity of the network that the Arduino can handle is a separate question, especially when it comes to training -- tens of thousands of iterations on training data. Training on a fast ...

10

Could you train a neural network on a microcontroller? Maybe, but please don't try. Could you use a NN for classification, etc on a microcontroller? Sure, as long as you can calculate the result of propagating the node and edge values and handle the multiplications.

10

From what I understand of your question, you'd like to know if inverse kinematics and reinforcement learning are trying to solve the same problem in the particular case of robotic manipulation. Of course both of these techniques can be applied outside of this particular realm, but let's focus on robot manipulation for now. You're right that inverse ...

10

For testing simple algorithms, you might be able to get by with a 2D simulator. There are a few out there that I am aware of: Stage: http://playerstage.sourceforge.net/index.php?src=stage STDR: http://stdr-simulator-ros-pkg.github.io/ Stage is an older, but useful, simulator which has integration with ROS (http: //wiki.ros.org/stage_ros) which will allow ...

9

One of the key measures of any machine learning algorithm is it's ability to generalize (i.e. apply what it has learned to previously unsceen scenarios). Reinforcement learners (RL) can generalize well but this ability is in part a function of the state-space formulation in my experience. This means that if you can find the right setup then the RL learner ...

9

There are a number of things to consider for your project. Since you are asking for the learning algorithms, I asume your hardware is or will be up and running. When getting your robot to learn, you should differentiate between on-line and off-line learning. Further, there is on-system and off-sytem learning, which can be combined with the previous category. ...

8

The function $f$ comes from the equation of motion for the inverted pendulum problem (inverted pendulum alone, not including the motion of the wheeled platform). If you consider your figure but ignore the side-to-side motion of the cart, then the equilibrium of moments about the hinge is: $\sum M = m g l \sin \theta - b\frac{d \theta}{dt}$ Where $m$ is ...

7

There is not a specific set of learning algorithms that you will need to implement. Genetic algorithms (GA), neural networks (GA), and reinforcement learning (RL) have all successfully been applied to the problem of gait generation. I can also conceive of ways to use unsupervised learning methods to approach this problem but I can't say for certain whether ...

6

"Optimal learning" is a very vague term, and it is completely dependent on the specific problem you're working on. The term you're looking for is "overfitting": (The green line is the error in predicting the result on the training data, the purple line the quality of the model, and the red line is the error of the learned model being used "in production") ...

6

Yes. If you only run it in feed-forward mode and do your training off-line somewhere else: I programmed a 3-layer (5-5-2) feedforward ANN on an Arduino UNO. It ran on a mobile robot. Whenever the robot would hit something, it would re-train the network. The feedforward portion of the net ran in real-time; while the back-propagation training took on the ...

5

Here's a paper that seems relevant: Policy Gradient Reinforcement Learning for Fast Quadrupedal Locomotion. Abstract: This paper presents a machine learning approach to optimizing a quadrupedal trot gait for forward speed. Given a parameterized walk designed for a specific robot, we propose using a form of policy gradient reinforcement learning to ...

5

Short answer: the strongest reinforcement effect comes from delivering a valuable reward on an intermittent (random) schedule. Longer version: One aspect of your question is about operant conditioning, at least as it applies to teaching maths to a complex organism. Applying this to machine learning is known as reinforcement learning. Economics (as per ...

5

Perhaps the best way to get started on this kind of problem is to take relevant coursework(either online or in real life) or to read an introductory book on this topic. A good introductory book on motion planning and SLAM is Principles of Robotic Motion. A good course on SLAM/Mobile Robots: Control of Mobile Robots

5

First, here's what you CAN do with those sensors. Assuming you are not constantly accelerating you can use the accelerometer to know which direction is "down" (the gyro can be used as well for faster updates). If there aren't any magnetic field disturbances, you can also use the compass to know which direction is forward. Usually this is done using either an ...

4

My understanding of your problem is that you would like to discover and navigate a 2D maze of irregular obstacles with a non-holonomic robot using a single forward-looking ultrasonic range sensor and wheel odometry. This is a hard problem. "Best" solution Although a "best" or "optimal" solution to this problem possibly could be implemented on an 8-bit ...

4

The mechanics of your vehicle are not extremely relevant here; I will assume that the motion your vehicle induces on the sensors will be within their specifications. Entire volumes have been written on "sensor fusion", which is the act of combining measurements from multiple sensors (e.g. your gyro, accelerometer, and compass). Doing this 100% accurately ...

3

These issues are addressed, to some extent, by the study of utility functions in economics. A utility function expresses effective or perceived values of one thing in terms of another. (While the curves shown in the question are reward functions and express how much reward will be tendered for various performance levels, similar-looking utility functions ...

3

You might want to have a look at my maze solving robot solution (http://www.benaxelrod.com/robots/maze/index.html). I used a Lego RCX which is more powerful than an 8bit microcontroller, but is still pretty resource constrained. I abstracted away most of the hardware problems to focus on the algorithm. It uses a flood-fill or A* type algorithm.

3

Unfortunately, I have no experience with ez-b, but I have looked over the site a little bit. I do, however, have lots of Arduino experience. The program is, indeed, stored on the board's local memory. However, it is very possible to write a program that can interact with your computer. With my Arduino, I often write programs that communicate with my computer ...

3

Gazebo is a good tool for what you want to do. Since you're using a custom robot you will need to build a model for the simulator to use. They have managed to make doing so pretty easy but for a quadraped I can imagine it will take a bit of time. Gazebo is also nice because it works well with ROS which means that if you build you could build a program to ...

3

This reminds me of QWOP (seriously). You need to plan a sequence of button presses to move forward. For every situation the runner finds himself in, we need to know what button to press, how long to hold it, etc. When you play that, you probably look a lot like this video I stumbled across (pun alert): NN training bi-pedal walkers What you need to ...

3

Yes indeed, it's possible to embed neural network in microcontrollers. There are many such examples of this in the scientific literature but I can cite a striking example of what can be done with a very simple MCU if you're smart enough. In Evolutionary Bits'n'Spikes, the authors describe the implementation of a real time spiking neural network AND a genetic ...

3

You can theoretically use just an accelerometer for determining motion, but it may not be accurate enough to achieve your goals. The big problem with accelerometers is drift over time (i.e., errors in the acceleration measurement get integrated twice), so your position accuracy significantly decreases over time. The severity of this problem depends on the ...

3

It actually makes sense that the dot product in both cases is the same (zero) because the dot product of two vectors does not consider the vectors' origins. Or in other words the math for the dot product places the two vectors at the same origin. In this sense there is no way to distinguish converging or diverging vectors. I think what you need to do is to ...

3

Regarding methodologies and tools, I recommend Chris Eliasmith's How to Build a Brain. It presents the Semantic Pointer Architecture (SPA), a cognitive model that has been realized in the open source Nengo toolkit. I have read the book's introduction and some of Eliasmith's papers, and so far the approach looks very promising.

3

You've asked more than one question, so I'll try to answer them in order. The Robotics community has not yet hit the limits of current hardware, so very little work is being done on the exotic cutting edge like neuromorphic hardware. The exception to this is software neural nets, which have come in and out of fashion for decades, and the Nv artificial ...

3

I think the problem you're going to find is that machine learning requires learning. If the goals or objective vary just a little, then manually adjusting the software shouldn't be too difficult (if you programmed it well), and it might only take a few trials for the AI to adjust to the new scenario. If the objectives change so much that you're looking at "...

3

Sure, a drone can land on a powerline. That's a standard task like the "peg in hole problem" for robotarms. The aim is to maneuver a UAV near to a highvoltage line and eating all the energy. The earliest paper was written in 2009 and has a nice plotchart of the Electro-Magnetic Field on page 8 Powerline perching with a fixed-wing UAV for directing the UAV in ...

3

I think yes it can but how? My options are here: Static system for conventional systems It should stand or hang to line/pole/special place like birds. Please watch video for an example: https://www.youtube.com/watch?v=MvRTALJp8DM Dynamic system Harmless/secure distance present day wireless charging methods: use low energy harvesting drone or an ...

3

I read a bit more and realized that in RL states and rewards accept a wide variety of interpretations and this is the real complexity nowadays of this learning problem. In case of PID values, problem can be formulated as the following: imagine a Kp value, it represents a state. Next state could be increase or decrease 0.1. Same with the next state, and so ...

Only top voted, non community-wiki answers of a minimum length are eligible