I know you can mod servos for feedback signals idk if that is something that may help the adaptive response. Adafruit makes a feedback servo kit that remembers your movements that you adjust by hand and will replay them. So basically if you grab a wheel and turn it back and forth for 3 seconds, you can press a loop button and it will replay your inputs. And all you have to do is solder a wire to the white wire inside the servo so that it transmits back to whatever input you send it to. (Feedback loop) it also sends precise signals so that the input is exactly where it was in servo positions. Maybe it’s worth looking into.
That is interesting, I actually have a spare that I can take apart and have a look. One of the issues I had in PT3 was being able to know where the servo was versus where it was being commanded to travel to. Timing is really difficult in these kind of of dynamic systems.
You could probably use a 6 axis board to keep the level like a gimbal. Maybe set an angle limiter to fight g force and improve stability. That would be super cool for rock crawlers to prevent rollover’s.
I do have an MPU6050 in this version, but it was mainly for data logging. In Pt3 I tried a groundhook controller scheme with a 9 axis MPU but it didn't work out. Building a test rig right now to isolate a single wheel and see if I can break it down better.
It is the future. Hell a good sized snap-on tool box will set you back 25 to 35 k why not have an organized closet with its own stacker bot that retrieves what ya need
Adding some roll unevenly (twist) will drastically change the handling (over/understeer). I think it would be really interesting to test just adding a bit of twist and seeing the effect on handling. By varying the twist, you would be able to control under / oversteer. I suspect you are unintentionally introducing twist in with your active lean on steering, resulting in your oversteer. Great work!
That's an interesting idea to have the uneven twist. I'm also curious about weight transfer as well. Weirdly, most of the papers on active suspensions focus on the response to bumps, and not actively enhancing the handling.
@@IndeterminateDesign Agreed on papers focusing on bump response, but I think that is because it looks like traditional control performance measures like disturbance rejection. Cars handling and controls don't overlap in lots of places, so I'm not too surprised that you don't see it more. I look at what you have as automated suspension ride height adjustment, so anything that racers can do with ride height, per wheel, you can do in real time, which is quite a game changer. Also, leaning into corners preserves your suspension travel, so you can avoid bottoming out in hard corners...
7:48 For this particular usecase (and not what what you were doing in the future), the measurment of the lateral acc can be much simpler. (You probably were misguided by papers doing dead reckoning and others, that need instantaneous lateral acc) You could have done vector addition on 3 axis. Then average the result over time. The result minus 9.81 is lateral acc. The longer dataset, the more precise result...
I'm going to look into this more. To make things simpler I'm working with a test rig for a single wheel suspension so I can eliminate some of the math involved. The question I always have had though, is for a 4 wheel active suspension, am I measuring the sprung mass velocity at the center of gravity, or relative to each wheel. Basically is it orientation depedent?
@@IndeterminateDesign my comment was in regards to measuring lateral acceleration (aka grip) of the whole car. I think your question is about the control now. I think you need to look at individual wheel motion for the fast control loop. I would use bandpass filter on position of each wheel. Low-pass portion is so you don't compensate for high frequency with those servos, as it may cause resonance. The high-pass portion of the filter is to remove any DC component (continuous lateral acceleration, or being on a slope). Data from accelerometer could be used for slow loop, (added on top of fast loop). So that could be lateral, roll, pitch acceleration. Pitch and roll give you combinations of corner pair acceleration. They could be compensated by applying more or less fierce, temporarily. Initial gain could be derived from momemnt of inertia and distance from the center. Divided by 10 to not overcompensate. Lateral acceleration could be used for tilting the car (more appropriate than steering angle, you don't want to tilt the car when drifting on ice, but maybe in practice feed forward based on steering is better). Again, you don't need instantaneous lateral acceleration, so it's just moving average of x+y+z (Vector addition) - 9.81. Generally output of both loops would be position, which we can assume to be proportional to force, thanks to the spring. Servo velocity may matter too due to the damper, but it is probably inconsequential. The test rig would be nice to isolate different components and frequencies !
Great concept. Please allow me one proposal: The force of the shock absorber should not last directly on the servo horn with respect to reliability and energy efficience, I feel there are kinematic possibities to adjust the shock absorber base with an indirect actuation.
That’s a great point. I’ve been thinking about this. The servo horns are quite strong and supported by double bearings. I’m thinking of adding some additional tension springs on each servo that remove the majority of the car’s static weight. Ideally, I will eventually move to a more sophisticated actuator like a voice coil motor or hydraulics.
@@IndeterminateDesign A servo always draws current when there is a moment on the servo horn. At least the servo horn could be designed in such a way that no projected lever arm acts in the end positions. That would be an easy solution. Or possibly one could arrange the adjustment mechanism at right angles to the direction of force, that would be more difficult and I would have to think for a long time ...
Great project, I'm sorry but some of the more technical stuff goes straight over my head lol. I've been wanting to try something like this on a rock crawler. Main thing I've struggled with is how I would mount and connect the servos to the shocks. Did wonder about mounting them direct, not sure whether the servos will be strong enough though. Does the cantilever design you've done on the rear shocks reduce the stress on the servos at all?
The rockers in the rear reduce the force the servo sees by about 50%. Ideally I would have them in the front as well if I could fit them. The servos seem very strong, but I may add some counter springs to take a little of the static weight off them and reduce the amp draw a bit. Check out a guy named Superscale 2020 on TH-cam. He sells a suspension setup for rock crawlers to make them move more realistically, and he directly connected his servos to the control arms.
2 ปีที่แล้ว
Can that oversteering be partialy caused by the extra weight that the car has in a hight position?
That is definitely possible, the center of gravity is much higher. I had the ability to turn on and off the active suspension control and I should have done more back to back testing that way.
I go over this in part 1, but high level the servos aren’t fast enough to dampen high frequency oscillations. So the shocks which are already quite soft will act to dampen the high frequency and the servos will focus on the lower wheel hop frequencies. It also lessens servo shock. Ideally, someday I’ll create faster actuators but those will be $$$.
Right now it’s just keeping the chassis level which does improve cornering. I’m working on a controller now that will actually respond to bumps (each suspension arm has an encoder). This is mainly a learning experiment to learn about different active suspension technologies.
Yes, that is the plan. The servos won’t handle all of the bumps because they’re too slow, hence the inline shocks. Someday I may try faster/lighter actuators if this proves to be effective.
Specifically for your suspension controller algorithm ;) Maybe use a A1324LUA-T Linear Hall Sensor - check rs-components online. (its kind of equivalent to a 10K pot depending on how you use it - took me ages to find and use for another project) Do some testing with some 6x3mm or 8x3mm neodymium mini magnets. Consider creating a tuneable delta offset in your algorithm to the Cant wait to see what you come up with! All the best!
![[Live] : ONE 169 วันนี้!! "อนาโตลี vs อูมาร์"](http://i.ytimg.com/vi/IK3YZVWuFJg/mqdefault.jpg)
SUper super cool project. The intro already had me hooked! I want to replicate something similar but with aerodynamic surfaces.
Wow, next level stuff. Much respect.
All key developments to build life-sized Batman Tumbler are now in place. Let's get going!🦇
I know you can mod servos for feedback signals idk if that is something that may help the adaptive response. Adafruit makes a feedback servo kit that remembers your movements that you adjust by hand and will replay them. So basically if you grab a wheel and turn it back and forth for 3 seconds, you can press a loop button and it will replay your inputs. And all you have to do is solder a wire to the white wire inside the servo so that it transmits back to whatever input you send it to. (Feedback loop) it also sends precise signals so that the input is exactly where it was in servo positions. Maybe it’s worth looking into.
That is interesting, I actually have a spare that I can take apart and have a look. One of the issues I had in PT3 was being able to know where the servo was versus where it was being commanded to travel to. Timing is really difficult in these kind of of dynamic systems.
This is great! I was looking to implement something similar, it's very cool to see this project.
Great project love your thinking keep at we all need this on our rc cars 👍👍👍
You could probably use a 6 axis board to keep the level like a gimbal. Maybe set an angle limiter to fight g force and improve stability. That would be super cool for rock crawlers to prevent rollover’s.
I do have an MPU6050 in this version, but it was mainly for data logging. In Pt3 I tried a groundhook controller scheme with a 9 axis MPU but it didn't work out. Building a test rig right now to isolate a single wheel and see if I can break it down better.
Leaning into the corners has an effect like climbing a hill, and you can get some very tight turning circles like in motogymkhana
Next a jonny 5 tool gofer for my maintenance shop . To go get tools and bring them from tool box room to the floor
It is the future. Hell a good sized snap-on tool box will set you back 25 to 35 k why not have an organized closet with its own stacker bot that retrieves what ya need
Love your work, thanks for sharing
Thanks so much!
Adding some roll unevenly (twist) will drastically change the handling (over/understeer). I think it would be really interesting to test just adding a bit of twist and seeing the effect on handling. By varying the twist, you would be able to control under / oversteer. I suspect you are unintentionally introducing twist in with your active lean on steering, resulting in your oversteer. Great work!
That's an interesting idea to have the uneven twist. I'm also curious about weight transfer as well. Weirdly, most of the papers on active suspensions focus on the response to bumps, and not actively enhancing the handling.
@@IndeterminateDesign Agreed on papers focusing on bump response, but I think that is because it looks like traditional control performance measures like disturbance rejection. Cars handling and controls don't overlap in lots of places, so I'm not too surprised that you don't see it more. I look at what you have as automated suspension ride height adjustment, so anything that racers can do with ride height, per wheel, you can do in real time, which is quite a game changer. Also, leaning into corners preserves your suspension travel, so you can avoid bottoming out in hard corners...
So cool!
7:48 For this particular usecase (and not what what you were doing in the future), the measurment of the lateral acc can be much simpler. (You probably were misguided by papers doing dead reckoning and others, that need instantaneous lateral acc)
You could have done vector addition on 3 axis.
Then average the result over time. The result minus 9.81 is lateral acc.
The longer dataset, the more precise result...
I'm going to look into this more. To make things simpler I'm working with a test rig for a single wheel suspension so I can eliminate some of the math involved. The question I always have had though, is for a 4 wheel active suspension, am I measuring the sprung mass velocity at the center of gravity, or relative to each wheel. Basically is it orientation depedent?
@@IndeterminateDesign my comment was in regards to measuring lateral acceleration (aka grip) of the whole car.
I think your question is about the control now.
I think you need to look at individual wheel motion for the fast control loop. I would use bandpass filter on position of each wheel. Low-pass portion is so you don't compensate for high frequency with those servos, as it may cause resonance. The high-pass portion of the filter is to remove any DC component (continuous lateral acceleration, or being on a slope).
Data from accelerometer could be used for slow loop, (added on top of fast loop). So that could be lateral, roll, pitch acceleration. Pitch and roll give you combinations of corner pair acceleration. They could be compensated by applying more or less fierce, temporarily. Initial gain could be derived from momemnt of inertia and distance from the center. Divided by 10 to not overcompensate.
Lateral acceleration could be used for tilting the car (more appropriate than steering angle, you don't want to tilt the car when drifting on ice, but maybe in practice feed forward based on steering is better).
Again, you don't need instantaneous lateral acceleration, so it's just moving average of x+y+z (Vector addition) - 9.81.
Generally output of both loops would be position, which we can assume to be proportional to force, thanks to the spring. Servo velocity may matter too due to the damper, but it is probably inconsequential.
The test rig would be nice to isolate different components and frequencies !
Very cool
Wow this is epic!!!
Great concept. Please allow me one proposal: The force of the shock absorber should not last directly on the servo horn with respect to reliability and energy efficience, I feel there are kinematic possibities to adjust the shock absorber base with an indirect actuation.
That’s a great point. I’ve been thinking about this. The servo horns are quite strong and supported by double bearings. I’m thinking of adding some additional tension springs on each servo that remove the majority of the car’s static weight.
Ideally, I will eventually move to a more sophisticated actuator like a voice coil motor or hydraulics.
@@IndeterminateDesign A servo always draws current when there is a moment on the servo horn. At least the servo horn could be designed in such a way that no projected lever arm acts in the end positions. That would be an easy solution. Or possibly one could arrange the adjustment mechanism at right angles to the direction of force, that would be more difficult and I would have to think for a long time ...
Should do a colab with Johnny q 90. He's making a sick car right now with a sweat motor
That would be cool, I love his videos. I think it will be a while before I can perfect this setup and make it small enough to fit in his car.
SO SWEET :)
Great project, I'm sorry but some of the more technical stuff goes straight over my head lol. I've been wanting to try something like this on a rock crawler. Main thing I've struggled with is how I would mount and connect the servos to the shocks. Did wonder about mounting them direct, not sure whether the servos will be strong enough though. Does the cantilever design you've done on the rear shocks reduce the stress on the servos at all?
The rockers in the rear reduce the force the servo sees by about 50%. Ideally I would have them in the front as well if I could fit them. The servos seem very strong, but I may add some counter springs to take a little of the static weight off them and reduce the amp draw a bit.
Check out a guy named Superscale 2020 on TH-cam. He sells a suspension setup for rock crawlers to make them move more realistically, and he directly connected his servos to the control arms.
Can that oversteering be partialy caused by the extra weight that the car has in a hight position?
That is definitely possible, the center of gravity is much higher. I had the ability to turn on and off the active suspension control and I should have done more back to back testing that way.
What RC car model are you using?
Why use shocks when the position of the arms are set by the servos? It seems that the shocks and servos are working against each other.
I go over this in part 1, but high level the servos aren’t fast enough to dampen high frequency oscillations. So the shocks which are already quite soft will act to dampen the high frequency and the servos will focus on the lower wheel hop frequencies. It also lessens servo shock. Ideally, someday I’ll create faster actuators but those will be $$$.
Very very nice ❗❗❗❗❗
Is that purpose to make car more stable wen cornering, or something else?
Right now it’s just keeping the chassis level which does improve cornering. I’m working on a controller now that will actually respond to bumps (each suspension arm has an encoder). This is mainly a learning experiment to learn about different active suspension technologies.
@@IndeterminateDesign oo i see, its mean all suspension independently to respon shock and bump and the respon controlling by some kind modul program?
Yes, that is the plan. The servos won’t handle all of the bumps because they’re too slow, hence the inline shocks. Someday I may try faster/lighter actuators if this proves to be effective.
Traxxas Hawk donor car?
Nope, RC10T.
@@IndeterminateDesign Thanks! I knew it looked familiar but couldn't place it.
Hall sensors are your friend
Specifically for your suspension controller algorithm ;)
Maybe use a A1324LUA-T Linear Hall Sensor - check rs-components online. (its kind of equivalent to a 10K pot depending on how you use it - took me ages to find and use for another project)
Do some testing with some 6x3mm or 8x3mm neodymium mini magnets.
Consider creating a tuneable delta offset in your algorithm to the
Cant wait to see what you come up with!
All the best!
Repect
Bruh my account got banned