Self-balancing electric bike

Valdior

1 mW
Joined
Jul 19, 2018
Messages
11
Hi Everyone!

On CES2017 Honda introduced self-balancing motorbike.
https://www.youtube.com/watch?v=hLZ6U71y700

BMW has their self-balancing plans too:
https://www.youtube.com/watch?v=NXMGi9kQ3TE

Yamaha didn't want to fall behind and also introduced self-balacing motorbike with very interesting balancing method:
https://www.youtube.com/watch?v=InTJW1TeCLs

I think you can see the trend of self-balancing/driving 2-wheel transport.
And what about e-bikes?
Would you want to have e-bike with self-balancing function (without heavy gyroscopes?)
It will look like normal e-bike (not heavy - circa 20 kg), but when you get off the bike it stends without help and can follow you (without load), so you can simply walk or stand and don't care about bike.
E-bike will have cameras to avoid pedestrians and walls.
Person will have bracelet, which bike will use as a beacon.
Wireless features to communicate with smartphone and all another stuff (smart home, V2X communication).
Also this e-bike will be able to have active anti-theft functions.
I think price could be like other e-bikes in this class (2-3k$)

Now i am working on balancing feature, and will be great to have some feedback from community.
Thanks!
 
I'm not quite sure what the big advantage would be. If you were really drunk or stoned and couldn't balance, you probably shouldn't be on a bike anyway. Not needing to put your feet down at a stop might be cool, but not really necessary. If it worked to keep you from crashing on loose gravel, snow or ice, that would be useful.
 
fechter said:
Not needing to put your feet down at a stop might be cool, but not really necessary.
FWIW, at least some places the law requires to put at least one foot down at a stop. Isn't enforced here AFAICT, but it is where Dogman lives, and a few others around the forum.


It does sound like an interesting project, and useful for the aged / infirm or others that have problems with balance, that would otherwise require a trike.
 
When I get on a bike it becomes self balancing.

I saw Hondas balancing demo at CES though and it looked like a neat novelty.
 
Sorry, maybe I didn't clarify that e-bike can't balancing with rider, it's physically impossible. Honda weights 200++ kg, is much heavier than rider, that's why Honda can balancing with rider.

when you get off the bike it stends without help and can follow you (without load), so you can simply walk or stand and don't care about bike.

The main idea is that you ride it like normal e-bike, pedal assist, but when you want or have to get off the bike you don't need to manage with bike - it becomes balancing when you release hands from the handlebar. So you can walk, talk on the phone, eat something, or shopping without need to manage your e-bike. Imagine self-parking feature=)

Maybe it's not so important to people who use e-bike only home-work-home, but for those who have e-bike life.
 
Well, I'm sure someone could use such a feature, but balancing with the rider or a load on it would probably be much more useful to more people.

Why would it be impossible for a bike that's lighter than the rider or load to balance with the rider or load on it?
 
amberwolf said:
Why would it be impossible for a bike that's lighter than the rider or load to balance with the rider or load on it?

Because of the height of the center of a gravity. WHen you weight 80kg and motorbike 200kg - center of a gravity will be much lower than with the e- bike is 20kg.

amberwolf said:
Well, I'm sure someone could use such a feature, but balancing with the rider or a load on it would probably be much more useful to more people.

Well, it is very interesting topic, and i think people themselves want to ride e-bikes.
For example, people want to have autopilots in cars because for most people car is just a way to go from A to B. And it's good when you can work or sleep when car takes the control.
But e-bikes, i think for many people it's a lifestyle, they love to ride bike, and our feature will only help them to enjoy e-bike. ( I mean people who not use bikes just home-work-home, as I mentioned before. With a popularity of e-scooters sharing it's not our case)
 
Just use a kickstand. I have seen people who are very into the bike life style, with clothing and nice bikes, but they still stop and fall over because they forgot their shoes were cleated into the pedals.

People who cannot ride a two wheeler because of balance issues have the option for three wheelers.
 
docw009 said:
Just use a kickstand. I have seen people who are very into the bike life style, with clothing and nice bikes, but they still stop and fall over because they forgot their shoes were cleated into the pedals.

People who cannot ride a two wheeler because of balance issues have the option for three wheelers.

Kickstand is a great idea, we will add it too.
E-bike will automatically use kickstand when parking needed.
(Yamaha has this function too)
 
Valdior said:
amberwolf said:
Why would it be impossible for a bike that's lighter than the rider or load to balance with the rider or load on it?

Because of the height of the center of a gravity. WHen you weight 80kg and motorbike 200kg - center of a gravity will be much lower than with the e- bike is 20kg.

amberwolf said:
Well, I'm sure someone could use such a feature, but balancing with the rider or a load on it would probably be much more useful to more people.

Well, it is very interesting topic, and i think people themselves want to ride e-bikes.
For example, people want to have autopilots in cars because for most people car is just a way to go from A to B. And it's good when you can work or sleep when car takes the control.
But e-bikes, i think for many people it's a lifestyle, they love to ride bike, and our feature will only help them to enjoy e-bike. ( I mean people who not use bikes just home-work-home, as I mentioned before. With a popularity of e-scooters sharing it's not our case)

Yep.E-bike can even go anywhere you like:work,home,exercise,walk the dog,picnic,etc.It can take you to somewhere which you couldn't go before.That is an amazing experience.
 
Maybe for those low-slung two wheel recumbents?

IMG_6962_.jpg
 
Yes. I already have this technology deployed on my bike.





OK. But in all seriousness the feature would be nice, nifty, and even useful at times. But I'm concerned about the added cost.
 
wturber said:
Yes. I already have this technology deployed on my bike.
:D

wturber said:
OK. But in all seriousness the feature would be nice, nifty, and even useful at times. But I'm concerned about the added cost.
Adding self-balancing function is not expensive, when you create e-bike from zero, we are aiming at a broad market.
 
i am adding autonomous capability to my bike (Duty Cycle) and i can say that self-balancing (while rolling) is possible. i have trained a CNN (Convolutional Neural Network) using Tensorflow.js on mobile and have it running in a game simulator. Right now I am only capturing/training on video, throttle, brake and steering. Theoretically, it would be possible to train on accelerometer and gyro datasets as well. The technique i am using is called behavioral cloning---it will predict steering angle based on inputs and generate a loss function based on actual human input. While not completely self-balancing, it would allow a person to ride hands-free (without all the complexity, energy and weight of a physical gyro).
 
j3tch1u said:
The technique i am using is called behavioral cloning---it will predict steering angle based on inputs and generate a loss function based on actual human input. While not completely self-balancing, it would allow a person to ride hands-free (without all the complexity, energy and weight of a physical gyro).
That's pretty cool. You could ride if you were super drunk.

It would be even better if it could somehow prevent common crash scenarios like hitting loose gravel in a corner. I've seen guys that could ride a bike on snow-packed ice without crashing while I could go about 5 feet before going down, so with the right skill it's possible.
 
j3tch1u said:
i am adding autonomous capability to my bike (Duty Cycle) and i can say that self-balancing (while rolling) is possible. i have trained a CNN (Convolutional Neural Network) using Tensorflow.js on mobile and have it running in a game simulator. Right now I am only capturing/training on video, throttle, brake and steering. Theoretically, it would be possible to train on accelerometer and gyro datasets as well. The technique i am using is called behavioral cloning---it will predict steering angle based on inputs and generate a loss function based on actual human input. While not completely self-balancing, it would allow a person to ride hands-free (without all the complexity, energy and weight of a physical gyro).

Very interesting! Maybe you can try to use less complex models instead of training heavy CNN?
One more thing - when you are drunk, riding on self-balancing bike in hands-free mode and suddenly bike decided to stop because of obstacle - you will fall down. You need to be ready to accelerate and decelerate or you will fall down.
 
my cnn is based on nvidia's model: https://devblogs.nvidia.com/deep-learning-self-driving-cars/
inference takes < 20ms (throttle, brake and steering) on google pixel xl (though i've only trained on a game sim).
i'm also using opencv in a webassembly worker to detect cars, lanes and orb features (pretty fast as well but not as robust as cnn and cannot utilize the gpu).

all in all, a simple cnn works really well to provide a steady stream of data for accelerating, steering, braking and optionally balancing. for proof of concept, i hope to be able to summon the bike and have it ride a short distance to me.

as for sudden stops and slips on gravel/snow, it will probably have to be set to kick in way before the point of no return and "ease" the bike back into a safe trajectory.

[pre]

buildModel() {
const model = tf.sequential();
model.add(tf.layers.conv2d({ filters:24, kernelSize:5, strides:2, activation:'elu', inputShape: [this.opt.imageHeight, this.opt.imageWidth, 3] }));
model.add(tf.layers.conv2d({ filters:36, kernelSize:5, strides:2, activation:'elu' }));
model.add(tf.layers.conv2d({ filters:48, kernelSize:5, strides:2, activation:'elu' }));
model.add(tf.layers.conv2d({ filters:64, kernelSize:3, strides:3, activation:'elu' }));
// model.add(tf.layers.conv2d({ filters:64, kernelSize:3, strides:3, activation:'elu' }));
model.add(tf.layers.dropout({ rate: this.opt.dropoutRate }));
model.add(tf.layers.flatten());
model.add(tf.layers.dense({ units:100, activation:'elu' }));
model.add(tf.layers.dense({ units:50, activation:'elu' }));
model.add(tf.layers.dense({ units:10, activation:'elu' }));
model.add(tf.layers.dense({ units:1 }));
return model;
}[/pre]


Valdior said:
j3tch1u said:
i am adding autonomous capability to my bike (Duty Cycle) and i can say that self-balancing (while rolling) is possible. i have trained a CNN (Convolutional Neural Network) using Tensorflow.js on mobile and have it running in a game simulator. Right now I am only capturing/training on video, throttle, brake and steering. Theoretically, it would be possible to train on accelerometer and gyro datasets as well. The technique i am using is called behavioral cloning---it will predict steering angle based on inputs and generate a loss function based on actual human input. While not completely self-balancing, it would allow a person to ride hands-free (without all the complexity, energy and weight of a physical gyro).

Very interesting! Maybe you can try to use less complex models instead of training heavy CNN?
One more thing - when you are drunk, riding on self-balancing bike in hands-free mode and suddenly bike decided to stop because of obstacle - you will fall down. You need to be ready to accelerate and decelerate or you will fall down.
 
I love it Ben!! Im excited to ride it!!
 
Hey!

We have some updates - now we are building prototype of self-balancing electric bike. Electronics is ready.
Steering wheel rotation mechanism is very challenging, because it should be small, compact, powerful, and not interfere riding. We have built one, but it's too slow, and we decided to build another variant.

We have a picture of the prototype (not product version!!!):
 
AltRiders said:
Looks really interesting, would love to get my hands on one to test.
Thank you, it's important for us.
As soon as we get the first prototype we will share it with community and some people wwill be able to make a test ride.
 
Hi people! After long silence, we can say, that we did a lot, made prototype and now we are testing self-balancing function.
Also we have a website with sketches and some feature descriptions. Video on the website is quite old.

https://rothembike.com/

Another our product which will be launched soon, is rear-view camera with radar for bicycles. It predicts collisions from behind and sends audible signals to the car, which can crash into your bicycle.

Have a good day and please leave comments!
 
If you want the demo video to be convincing, you should show the entire bike, not just the front wheel.
 
Back
Top