7 ms·

A motorbike-racing robot

62 points by groodt 9y ago

haihaibye 9y ago
Nice. Now we just need the robot to ride with 1 hand and reload a shotgun with the other.
- EGreg 9y ago
  Why not just have three hands? And why have a shotgun when a laser is better!
  - King-Aaron 9y ago
    I don't think James Cameron thought that far
  - taneq 9y ago
    It's an infiltration model designed to pass as human until it reaches its target.
- older 9y ago
  Then fight this one: https://www.youtube.com/watch?v=6LjKfs7MjYE https://www.youtube.com/watch?v=6LjKfs7MjYE
EGreg 9y ago
"We decided that physically attaching Motobot to the bike wasn’t cheating because its hands had to grip the handlebars" Come on guys. This is just like Watson's finger or AlphaGo's precomputation. It is cheating obviously!!
- rorykoehler 9y ago
  If it's cheating it's not very useful in gaining an advantage. Hanging the upper body out and forward is a core component of the contemporary winning motogp riding technique.
  - EGreg 9y ago
    Of course it's cheating. The humans have to deal first and foremost with not falling off. That is very costly. They do a lot of actions and calculations to prevent this. The robot has unfair advantages like not caring about getting hurt, and being assured it can't fall off. Given a human 100% sure of that I bet they can also go 200 km/h or even 400!
- visarga 9y ago
  > AlphaGo's precomputation What are you talking about, training the Go neural net?
- jlebrech 9y ago
  the robot should get back on the bike when it falls off it, and then you can tell it to experiment with leaning etc.. and just leave it for months on end.
atcole 9y ago
As a qualifier, missing by 30 seconds on a 85 second lap is not very close. But a 1:55 at Thunderhill would be consider a fairly decent amateur time on a 600cc bike so they are definitely making progress.
waiseristy 9y ago
As long as they have the robot sitting there static, they're going to be hard pressed to beat a human. There's a lot more going on on a bike than just fiddling with the throttle and steering
- asteli 9y ago
  The humanoid form is just for show. You need a handful of actuators to work the control and a way to power them. Throttle is already servo-actuated on the stock bike. The robot will have the advantage in mass and mass distribution. What I do think the bots will have a hard time doing is replicating the intuition that people like Rossi and Marquez have. Understanding what parts of the track have the most grip, braking/accelerating/cornering on the ragged edge of traction. And that's only half of what's happening in race conditions -- you have to contend with a grid full of other riders taking offensive lines and passing only inches away. A robot motorcycle might be able to take on GP-level riders at Thunder Hill in a time trial in a few years, but it's hard to imagine the same thing happening at, say, the Isle of Man TT, or anything approaching race conditions.
  - fyfy18 9y ago
    > you have to contend with a grid full of other riders taking offensive lines and passing only inches away It'll be interesting to see what happens if we have robots and humans racing together. On one hand they will be programmed to win, but on the other hand, harming humans is a big no no. It's likely that this will happen before we have true AI, so it'll be the human overlords (i.e. team mangers) who make that moral decision. Given some of the things racing teams have done before to win, it could be quite concerning...
- dazc 9y ago
  “Motobot had to be taught how to balance the bike at lean angles from zero to over 50 degrees, at speeds between 5km/h and over 200 km/h." Article doesn't say, specifically, but it must be doing something other than just sitting static to achieve this? [edit] https://www.youtube.com/watch?v=4dUb5T7DU-Q https://www.youtube.com/watch?v=4dUb5T7DU-Q Bike has outriggers!
Animats 9y ago
How soon will Doordash have this technology?
groodt 9y ago
It is using GPS for navigation at the moment. I wonder if the reduction in Lidar size and advancements in cameras would mean that in theory these things could be raced in groups. That would be cool.
netgusto 9y ago
A video of this robot: https://www.youtube.com/watch?v=BjZPvXKewFk https://www.youtube.com/watch?v=BjZPvXKewFk
vr46 9y ago
There's a lot more to racing than simply being able to go fast around a track solo, but this was a fabulous piece of work from Yamaha. This was last summer, I think, so it's a bit late for the Beeb to be reporting on it. Soon, I imagine that the bike will also start to be tailored to the robot in the way that it's tailored to a human (two humans, actually, Rossi and Viñales).
asteli 9y ago
> “Why a motorbike?” ponders Hiroshi Saijou. “Because it is very difficult to do, and it had never been done before.” I feel compelled to point out that it has in fact been done before. There was an autonomous motorcycle -- Ghostrider -- which was entered into the '04 and '05 DARPA Grand Challenge, and now rests in the Smithsonian. [1] [1] http://americanhistory.si.edu/collections/search/object/nmah_1332301 http://americanhistory.si.edu/collections/search/object/nmah...
- jstanley 9y ago
  That's a modified motorcycle. The one in the article is a normal motorcycle with a robot sitting on it.
  - asteli 9y ago
    Putting the actuators on the motorcycle is the easy part. From a control perspective, it doesn't much matter whether the actuators are packaged into a humanoid form or not.
    - jstanley 9y ago
      You have much more precise control over things like throttle position if you're controlling a fly-by-wire throttle electronically compared to actuating a physical throttle mechanically. In principle I agree with you, but I think the inaccuracies will stack up quite quickly.
    - vr46 9y ago
      Getting a motorcycle to change direction is more than just actuating a control, body position, centre of gravity and using the legs to push down on the pegs are all things racers do.
      - asteli 9y ago
        This is true! Nice username, btw! I think some of this comes down to how closely you have to mimic a human rider. Does the robot have to weigh as much as a plausible GP rider? If not, then the robot plainly has a mass advantage. Would that be counteracted by the robot's inability to shift mass and drag knee/elbow like a Rossi or Marquez can? I actually don't know.
Benjamin_Dobell 9y ago
Footage: https://global.yamaha-motor.com/showroom/motobot/ep3/index.html https://global.yamaha-motor.com/showroom/motobot/ep3/index.h...
pretty_dumm_guy 9y ago
Hello guys, I am finding it hard to understand how they managed to model the system(i.e bike). I am somewhat familiar with control theory(did some undergrad courses on them). But this system seems like a non-linear system and coming up with a mathematical model seems like a herculian task to me.(I once modeled a conical tank and it was not quite easy) Could someday provide me any idea about what they have done here ? How they modeled this system ? What control algorithms they used here ? Any pointers might be helpful as well. P.S - Haven't read about developments in control theory in quite some time now. So forgive me if my comment is quite basic. Thank you.

62 points by groodt 9y ago

haihaibye 9y ago

Nice. Now we just need the robot to ride with 1 hand and reload a shotgun with the other.

EGreg 9y ago
Why not just have three hands? And why have a shotgun when a laser is better!
- King-Aaron 9y ago
  I don't think James Cameron thought that far
- taneq 9y ago
  It's an infiltration model designed to pass as human until it reaches its target.
older 9y ago
Then fight this one: https://www.youtube.com/watch?v=6LjKfs7MjYE https://www.youtube.com/watch?v=6LjKfs7MjYE

EGreg 9y ago

"We decided that physically attaching Motobot to the bike wasn’t cheating because its hands had to grip the handlebars" Come on guys. This is just like Watson's finger or AlphaGo's precomputation. It is cheating obviously!!

rorykoehler 9y ago
If it's cheating it's not very useful in gaining an advantage. Hanging the upper body out and forward is a core component of the contemporary winning motogp riding technique.
- EGreg 9y ago
  Of course it's cheating. The humans have to deal first and foremost with not falling off. That is very costly. They do a lot of actions and calculations to prevent this. The robot has unfair advantages like not caring about getting hurt, and being assured it can't fall off. Given a human 100% sure of that I bet they can also go 200 km/h or even 400!
visarga 9y ago
> AlphaGo's precomputation What are you talking about, training the Go neural net?
jlebrech 9y ago
the robot should get back on the bike when it falls off it, and then you can tell it to experiment with leaning etc.. and just leave it for months on end.

atcole 9y ago

As a qualifier, missing by 30 seconds on a 85 second lap is not very close. But a 1:55 at Thunderhill would be consider a fairly decent amateur time on a 600cc bike so they are definitely making progress.

waiseristy 9y ago

As long as they have the robot sitting there static, they're going to be hard pressed to beat a human. There's a lot more going on on a bike than just fiddling with the throttle and steering

asteli 9y ago
The humanoid form is just for show. You need a handful of actuators to work the control and a way to power them. Throttle is already servo-actuated on the stock bike. The robot will have the advantage in mass and mass distribution. What I do think the bots will have a hard time doing is replicating the intuition that people like Rossi and Marquez have. Understanding what parts of the track have the most grip, braking/accelerating/cornering on the ragged edge of traction. And that's only half of what's happening in race conditions -- you have to contend with a grid full of other riders taking offensive lines and passing only inches away. A robot motorcycle might be able to take on GP-level riders at Thunder Hill in a time trial in a few years, but it's hard to imagine the same thing happening at, say, the Isle of Man TT, or anything approaching race conditions.
- fyfy18 9y ago
  > you have to contend with a grid full of other riders taking offensive lines and passing only inches away It'll be interesting to see what happens if we have robots and humans racing together. On one hand they will be programmed to win, but on the other hand, harming humans is a big no no. It's likely that this will happen before we have true AI, so it'll be the human overlords (i.e. team mangers) who make that moral decision. Given some of the things racing teams have done before to win, it could be quite concerning...
dazc 9y ago
“Motobot had to be taught how to balance the bike at lean angles from zero to over 50 degrees, at speeds between 5km/h and over 200 km/h." Article doesn't say, specifically, but it must be doing something other than just sitting static to achieve this? [edit] https://www.youtube.com/watch?v=4dUb5T7DU-Q https://www.youtube.com/watch?v=4dUb5T7DU-Q Bike has outriggers!

Animats 9y ago

How soon will Doordash have this technology?

groodt 9y ago

It is using GPS for navigation at the moment. I wonder if the reduction in Lidar size and advancements in cameras would mean that in theory these things could be raced in groups. That would be cool.

netgusto 9y ago

A video of this robot: https://www.youtube.com/watch?v=BjZPvXKewFk https://www.youtube.com/watch?v=BjZPvXKewFk

vr46 9y ago

There's a lot more to racing than simply being able to go fast around a track solo, but this was a fabulous piece of work from Yamaha. This was last summer, I think, so it's a bit late for the Beeb to be reporting on it. Soon, I imagine that the bike will also start to be tailored to the robot in the way that it's tailored to a human (two humans, actually, Rossi and Viñales).

asteli 9y ago

> “Why a motorbike?” ponders Hiroshi Saijou. “Because it is very difficult to do, and it had never been done before.” I feel compelled to point out that it has in fact been done before. There was an autonomous motorcycle -- Ghostrider -- which was entered into the '04 and '05 DARPA Grand Challenge, and now rests in the Smithsonian. [1] [1] http://americanhistory.si.edu/collections/search/object/nmah_1332301 http://americanhistory.si.edu/collections/search/object/nmah...

jstanley 9y ago
That's a modified motorcycle. The one in the article is a normal motorcycle with a robot sitting on it.
- asteli 9y ago
  Putting the actuators on the motorcycle is the easy part. From a control perspective, it doesn't much matter whether the actuators are packaged into a humanoid form or not.
  - jstanley 9y ago
    You have much more precise control over things like throttle position if you're controlling a fly-by-wire throttle electronically compared to actuating a physical throttle mechanically. In principle I agree with you, but I think the inaccuracies will stack up quite quickly.
  - vr46 9y ago
    Getting a motorcycle to change direction is more than just actuating a control, body position, centre of gravity and using the legs to push down on the pegs are all things racers do.
    - asteli 9y ago
      This is true! Nice username, btw! I think some of this comes down to how closely you have to mimic a human rider. Does the robot have to weigh as much as a plausible GP rider? If not, then the robot plainly has a mass advantage. Would that be counteracted by the robot's inability to shift mass and drag knee/elbow like a Rossi or Marquez can? I actually don't know.

Benjamin_Dobell 9y ago

Footage: https://global.yamaha-motor.com/showroom/motobot/ep3/index.html https://global.yamaha-motor.com/showroom/motobot/ep3/index.h...

pretty_dumm_guy 9y ago

Hello guys, I am finding it hard to understand how they managed to model the system(i.e bike). I am somewhat familiar with control theory(did some undergrad courses on them). But this system seems like a non-linear system and coming up with a mathematical model seems like a herculian task to me.(I once modeled a conical tank and it was not quite easy) Could someday provide me any idea about what they have done here ? How they modeled this system ? What control algorithms they used here ? Any pointers might be helpful as well. P.S - Haven't read about developments in control theory in quite some time now. So forgive me if my comment is quite basic. Thank you.