BJR_01

Project Logs

Collapse

BJR_LOG_14 Stable control
Tamas Feher • 08/26/2024 at 19:49 • 0 comments
Sorry for the radio silence, in the past couple months I've been busy with getting the software ready.
I will write a couple logs about the software later, but for now, please enjoy this demonstration of the control system:

It's jittery, it's wobbly, it's loud, it's noisy, it's inaccurate, but at least it's stable!
Currently it's a simple PD controller, but more on that in a follow up log.

I'm so happy with this milestone, but we are far from being done.
- The control needs to be polished up
- The Base PCB needs to be designed, as it's all on breadboard now.
- The signature middle cable need to be selected
- The Cladding would need a little more work
After that, we might have a viable MVP!
Looking forward for the next steps, stay tuned for updates by following this project.
Please give it a like as well, to keep me encouraged to continue
BJR_LOG_13 Resistive plate PCB
Tamas Feher • 06/12/2024 at 14:12 • 0 comments

Introduction

In this log we will take a look at the resistive plate PCB, see it's functionality, where it sits in the assembly, and how it was assembled. Lastly I'll go over the lessons learned.

Resistive plate PCB

This pcb serves 4 main functions:

1. It controls the resistive touchscreen through the TSC2046 chip

2. It serves as a strucutral support that holds the resistive touchscreen, and the adjacent elements.

3. Anchors the middle hub, which serves as the holding point of the universal joints

4. It houses a 6DoF IMU as well, to be able to tell which angle does the plate sit relative to the gravity vector.

CAD view of the resistive plate pcb

here you can see how the PCB contributes to the top assembly. (the green hexagonal part)

PCB assembly
The chosen IMU (LSM6DS3TR-C) had a leadless package, and a very fine pitch. I didn't feel comfortable soldering it by hand, so I tried out for the first time stencil soldering. At the end it turned out better than I expected it would, however I did have to touch up the connectors by hand. I recorded the process, so please enjoy this short video of the process:
Let me know in the comments if I should improve the soldering process.
Lessons learned
- The PCB manufacturer quotes the PCB based on size, so using the PCB as a large structural element might not been the best choice. In the future it's best to keep the PCB as small as possible, and use other material for support.
- The threaded anchors rip the copper off easily from the PCB, so I had to glue them in position, which helped at the end. It's a good lesson again not to use PCBs for structural support.
Summary
Over all the resistive plate PCB will be an important part of the build, and now I have a working prototype of it.
BJR_LOG_12 New renders
Tamas Feher • 05/18/2024 at 20:59 • 0 comments

I've made a couple more renders, that elaborates more on the visual identity I envisioned for the project.
Please take a look, and tell me what you think:
The bottom and the top parts are geometric and very clean looking in terms of visual aesthetics. In the meantime the middle layer is unapologeticly exposed, celebrating the raw robotics. It exposes the universal joints, the pistons and a striking cable that flexes as the plates moves around.
The renders were done in Blender by exporting the model from onshape. Here are a couple more setups:
There are still a lot to be designed here, the cable is just a placeholder and the bottom shell is to be detailed further, but this would be the general outline.
Tell me what's your idea to improve the visual identity here.
BJR_LOG_11 Reviving the project
Tamas Feher • 05/09/2024 at 20:14 • 0 comments

Introduction

The last log has been a couple months ago, In the meantime I moved houses and now that I'm settled in, let's get this project off the ground again. In this log I'll introduce an alternative design that uses resistive touch sensing instead of capacitive or inductive sensing.

The reason for the change

In the earlier logs it looked like the inductive sensing would be a suitable candidate for this project, so why is it changing? you might ask.

It's true, however it takes a fair amount of design, prototyping, coding to make it into a feature that can accurately detect the ball's location. There are many other subsystems that I need to prove out. If I spend so long to get this tech off the ground, the rest of the subsystems will stay at risk.

In the industry the metric for how mature a technology is is called the Technology readiness level.

Resistive touchscreen technology

In BJR_LOG_03 I've introduced the resistive touchscreen, but just for a recap, here is how it operates:

In a resistive touch screen, there are 2 conductive films separated by a thin layer of air gap. Once something applies pressure to the top film, the contact is made by the 2 conductive surfaces. opposing edges of the films are connected to electrodes, so if a voltage is applied across them, with the contact point it forms a voltage divider. The voltage can be measured and from there the touch location can be interpreted.

The reason why I chose this over the other technologies is it's ease of use. Resistive touchscreens are available on Mouser, they are relatively cheap, and there is a wide support for controlling them. I chose the TSC2046 IC for control.

TSC2046 Rust driver

I chose rust as the main programming language for this project, simply because I enjoy writing firmware in this language. I published a driver for the TSC2046 chip on crates.io. If you ever happen to need it.

Mechanical assembly

Now that the plate is not a single pcb, but rather a combination of multiple elements, I needed to redesign a little the top part of the robot. While I was at it, I thought I would give it a little personality as well.

Summary
The development continues with a resistive touchscreen as the ball detection device.
BJR_LOG_10 Designing the inductive sensor
Tamas Feher • 01/30/2024 at 18:22 • 0 comments

Introduction

In the last log, we tried using Capacitive sensing to sense the ball, but that ultimately failed.
In this log, we are going to explore an inductive sensing method to detect the ball's position on the plate.

Inductive proximity sensing

There are multiple ways to exploit inductance for proximity sensing. In this article, we will focus on the resonant sensing principle.

The sensing inductive coil and a capacitor are connected in parallel to form an LC tank circuit. The inherent resonant frequency of this circuit can be calculated as

As a conductive target approaches the inductive coil, eddy currents form on the surface of the conductive target. The magnetic field of these eddy currents resists the current of the inductive coil, which reduces the inductance of the system and increases the resonant sensing frequency.

Here is an example LC circuit with 2H of inductance

And the same example but with half the inductance

You can see that the frequency of resonance has changed and this is the property that we will measure in this experiment. You can find out more about the principle of operation in this

application report from TI.

For testing, I'm using the TI LDC1614 high-resolution inductance to digital converter.

Detecting 2d position with inductive sensing

The way I chose to construct the sensing elements is by lining up multiple sensing coils next to each other. The sensing elements are spiral tracks printed on a PCB. They've been elongated into a "Horse race track" shape, so that they would be sensitive to the ball's movement only in one axis. Later tests confirmed that moving the ball alongside the racetrack doesn't influence the measurement to a significant degree.

Multiple of these tracks have been put next to each other and interlaced in different directions on a 4 layer pcb, so that we have sensing resolution in both the X and Y directions.

The final test pattern looks like a weaving pattern:

This way the vertical racetracks detect the X position by observing which coil has changed it's inductance the most. Hopefully to sub mm accuracy. Same applies to the horizontal racetracks.

Designing the PCB

The issue with these racetracks is that there is no built in feature in KiCad (Open source electronics design software) to lay spirals out. Doing it by hand would be a grudgingly long job. Instead I opted of automating the process through python scripting.

KiCad's python scripting API is a bit difficult to use due to lack of official documentation, so I opted to offload the racetrack making into a dedicated python script, that outputs a JSON file which contains the details for all the tracks that it needs to contain. The plugin I wrote for this job takes in that json file, and imports all the primitives into KiCad. I wrote a visualizer, that enabled me to iterate on the generator without spamming KiCad with failed tracks.

Here are the results:

Testing the Sensitivity

The big question about this side route is if we are going to be able to make enough change in the measured inductance that we can measure it well enough? Given that we have pcb coils, interviewing traces, and spherical target, instead of a flat one.
To test this I used the LDC1614EVM evaluation board and mounted the ball on a plastic rod, so it wouldn't interfere with the measurements. The plastic rod was fastened to an X-Y positioning table, that let me set the ball's position with sub mm accuracy.

After painstakingly taking a measurement at every single millimetre and manually recording the measured value, here is what we got:

The results are promising. We are seeing a 100000 counts maximum activation on the Y axis channels, and around 75000 counts maximum activation on the X axis. With this level of resolution, i'm happy to continue exploring in this direction

Summary

In this log we've introduced inductive...
Read more »
BJR_LOG_09 Testing the capacitive touch plate
Tamas Feher • 01/14/2024 at 17:11 • 0 comments

Introduction

In BJR_LOG_08, we've gone through how a capacitive touch-capable PCB was designed and manufactured. In this log, we will look over how this was tested and the results of the test for use in this project.

Test setup

The Capacitive touch plate exposes 2 connectors. One for the main controller, which will be used in normal operation, and a debug connector designed for testing the plate.

For testing, we will bypass the central controller altogether, and connect the plate to the computer through a USB bridge. The bridge is a PROMATE D21 . It's the cheapest bridge I could find at £78.22.

The reason I'm willing to spend on a device just to verify the operation of something that will be controlled through other means is the number of unknown factors. Let's say I download the I2C device driver for the MaxTouch chip, and start communicating with it through the central controller. If anything goes wrong, I have little idea whether my PCB design went wrong, my I2C bus is busted, the device driver is buggy, or my default configuration is flawed. The fewer unknown factors I expose for a test, the sooner I can conclude. By using an off-the-shelf bridge, and their tuning software, I can eliminate most of the unknown factors.

Test Method

First I tested the 3.3V bus. Applying the right voltage through a test point and see if it's using the right amount of current. Then went onto the 5V source and repeated the test. Once I was confident the board wouldn't burst into flames if I connected it to the bridge I went ahead and prayed it would recognise my board. By some divine miracle, it did! No touch-ups were needed, the connection was straight-up established!

Testing Touch capability

First I tested how well it detected human touch. The measurement was all over the place at the beginning. The reason for that is the mutual capacitance sensor's response to proximity. At first, the capacitance starts slowly decreasing. As the proximity reduces to near zero, the capacitance starts shooting up rapidly as seen in the graph below:

You can learn more about the physics of this phenomena in this youtube video, where I got the graph from.
The sensor only operates in the "dip" region, so to be able to detect my finger I needed to add a couple of layers of non-conducting material between the plate and my finger. (tape in my case for fine-tuning). Once the right dielectric thickness was found, the plate was working quite well to detect my finger. With a bit more work, I could have turned this into a decent touch panel.

The next video shows on a surface plot what the plate recognizes from my finger:

In the next video, you will see that the response is still great even if I hold a bearing ball in my hand.

|f I'm wearing thick gloves, there is still some response, but not strong enough to be able to detect a singular touch point, so the internal filter just gives up.

Now here comes the interesting bit. If I grab the steel ball with my gloved hand, and roll it around on the sensing surface, there is still too little activation of the sensor to pick up a touch signal.

Why is that? After looking into it I found this: It's not enough for an object to be conductive to be detected by the capacitive sensor, but also needs to be able to disrupt the electrostatic field well enough so that the capacitance change is picked up by the receiver. To disrupt the field, the object would need to be either coupled to the ground or be large enough to make a dent in the electrostatic field. Sadly the bearing ball I planned to use achieves neither. If I increase the size of the ball, the signal gets better, but with the weight limit of the actuators I can't increase it large enough to be usable. This taps into the study of tangibles on capacitive touch screens. Many have tried, but most of them came to the same conclusion: It would need a human to touch the device to be...
Read more »
BJR_LOG_08 Capacitive sensing plate
Tamas Feher • 12/18/2023 at 18:38 • 0 comments
Exploring Capacitive Touch

Introduction

In this log I'll introduce my first attempt at sensing the ball's X,Y position on a plate, using Capacitive touch sensing. The entire solution is self contained within a single custom PCB

The Appeal of Capacitive Touch

In the log BJR_LOG_03 Plate subsystem breakdown I introduced the main touch sensing technologies I considered, and landed on Capacitive touch sensing. Just to recap, it operated on the basis of detecting the change in capacitance in the panel when a finger/stylus touches the surface. This technology is used in almost all smartphones, and laptop touch pads. In our use case the sensor doesn't need to be transparent, so it can be all self contained within a single PCB.

The Design

The circuit itself if fairly simple. Here is the high level overview:

I threw in an imu into the design as well to be able to compensate for uneven floor. It might not be needed in the long run, and it doesn't hurt the prototype. The debug connector helps connecting the touch controller to a computer without having to write a driver for the device.

I used KiCad to create a prototype PCB that could be tested with the system. here is the schematics of it:

The main star of the assembly is the ATMXT336UD. It's a touch screen controller with 336 nodes from the maXTouch product family. It does all the measurement, signal processing, filtering, the whole bundle. It reports touch information on an I2C bus.

The sensing elements are arranged into rows and columns. The Columns are driven, and cycled through one by one. The rows are measured against the active column of sensing elements. This is how the 2d plane is covered with so many nodes.

A 4 layer pcb was used, where the layers look like the following:
1. Layer: The top most layer contains the sensor nodes. A diamond pattern was used, to maximize sensitivity of the mutual capacitance sensor.
2. Layer: Contains the interconnection between the column nodes:
3. Layer: Creates the shielding for the sensors. A soling ground plane fill would disrupt the electrostatic field, so a hatched fill is used with 25% infill.
4. Layer: Is the component and signal layer.
Some components had way too narrow pitch for me to comfortably hand solder it, so I asked the PCB manufacturer to populate the boards for me.

Here are some pictures of the complete board:

The sensing area is much smaller than the entire plate. The application note recommended to use rectangular footprint.

The next picture shows the diamonds sensing pattern.

Resources

This board was very much out of my comfort zone, as I've never used capacitive sensing before. Thankfully there are great resources available online which holds your hand through the design process with examples and application notes. The most useful resources for me were:
- CapTIvate™ Technology Guide Gives a good general overview of capacitive sensing methods and applying it to TI's capacitive inputs.
- Design Guidelines for PCB Layouts for maXTouch Controllers Helped with the layout of the components and sensing elements.
- QTAN0080: maXTouch Sensor Design Guide Gave help to tailor the design to the maXTouch product range.
Results

The next log will deep dive into the evaluation of this board. To give a bit of spoilers, the board Functionally works well, it detects human touch very well, however the touch of the ball yields very inconsistent results. We will see why in the next log.
Make sure to follow and like the project if you're interested how the plate operates in the target environment.

Takeaways

I hope this log helps show that an unfamiliar technology can be safely explored through manufacturers' application notes.
BJR_LOG_07 Central Control Circuit Block Diagram
Tamas Feher • 12/02/2023 at 17:34 • 0 comments

Let's look at the central control circuitry next.

If we take a look at the subsystem breakdown in BJR_LOG_02, we can see the central control circuit already has it's inputs and outputs defined. The following figure is a context diagram, which shows the Control Circuit and it's interfacing blocks.

Let's go over the interfaces and form them into requirements:

1. The Central Control Circuit shall be powered by 24VDC power

2. The Central Control Circuit shall control 3 stepper motors with 0.6A current per winding

3. The Central Control Circuit shall listen to the input of 3 endstop NO (normally open) switches mounted at the end of the actuators.

4. The Central Control Circuit shall listen to the input a user button

5. The Central Control Circuit shall communicate with the plate through I2C protocol (It hasn't been established yet how the plate will communicate with the base, a later log will go into more details on the why and the hows.)

The main component that controls the stepper motors is the TMC2209. It has been selected for it's StealthChop2 and CoolStep features which provides noise free operation at lower speeds and reduced current use in idle mode, which helps with heating of the actuators.

The datasheet helps identifying the main interfaces of this device, which are:
- Motor power in: 24VDC in our case

- Logic power in: 3.3VDC in our case
- Step/Dir: traditional stepper motor driver interface, where the motor executes a step every time a falling edge is detected on the Step line. The direction of the step is determined by the Dir input.

- EN: shorts the motor coils, and disables the output. It's useful for the off state and initialisation.

- Diag: pin offers feedback to the controller if anything went wrong during operation.

- UART: is a one pin serial interface to set the drivers internal registers.

Now we have all the necessary components to devise a block diagram for the circuit:

In this circuit a single microcontroller hosts all the computation needed. Based on the block diagram it needs to be capable of hosting:

- 9 Digital outputs

- 10 Digital inputs

- 3 UART channels

- An I2C bus

- SWD programming

Given these requirements I chose the STM32F446RE microcontroller.

There is no elaborate reason why I chose this one, I just had a couple of these lying around in a drawer.

With this the Circuit is ready prototyping and the software to be started.
If you're interested in the project, give it a like, and leave a comment on what part you would like to see next.
BJR_LOG_06 First assembly
Tamas Feher • 11/21/2023 at 18:01 • 0 comments
The biggest risk on the mechanical side of the project is backlash and flex in the assembly.

Backlash is difficult to account for in control systems so it's better to tackle it at the source.

To make the build I SLA resin printed all non standard parts, here you can see the results:

There were no major issues with assembly, however there were some issues with backlash and flex. On the video you can see me stressing the assembly to see where the weak points are.

I found the following issues:
- The base of the assembly is not too sturdy, introducing a lot of flexing
- The axle of the actuator mount is an M2 threaded rod, which has a slightly smaller outer diameter than 2mm (0.079 inches).
- The M2 axle is not particularly rigid, so it introduces some amount of flex as well.
- The plate holding bracket has some axial play introducing some backlash.
Other observations:
- The actuator could be mounted 180 degrees around, so that cabling wouldn't need to go around it to meet in the middle.
- It's pretty and I'm very happy with it
These issues will be tackled in the next version (BJR_02) but it's not severe enough to stop in my tracks to continue.
BJR_LOG_05 Actuator trade study
Tamas Feher • 11/19/2023 at 15:41 • 1 comment
Now there are enough requirements available to decide on the technology used for the linear actuators. Let's recap what are the relevant requirements:
- The BJR shall be silent during the operation
- The stroke length of the actuator shall be at the minimum of 13.5 mm ( 0.5315 inches)
- The output force of the actuator shall be at the minimum of 4.9N (1.101 pounds)
- The piston speed shall be at at the minimum 23.8 mm/s (0.937 in/s)
Overall this project needs actuators that are fairly quick with low force output. This gives an ideal candidate to direct drive motors.

Direct drive means no gear reduction involved in the drive train. It's beneficial in the sense as well that gear inefficiencies usually add noise to the system, while this project is aimed to be silent.

it's not enough to analyze the force and speed separately, as the actuator needs to be able to produce the prescribed force at the given speed.

The figure above shows that even though a motor is capable of delivering the needed speed and force, it might not be suitable for the application.

Let's see what technologies we would run up against each other in this trade study:

Linear motors

Linear motor technology has recently evolved a new form of motors, where the magnets are contained within a shaft, and the motor only contains the windings.

This technology enables very high-speed linear motion. The downside has been the technology did not break into the market yet, and the prices are still quite high. I was quoted over £400 (~498 USD) for a single unit.

Captive linear stepper motors

That's quite a mouthful, so let's break down how these elements describe the technology.

A stepper motor is a type of motor that moves in precise, fixed steps in response to electrical signals. Linear means that the rotation is converted to linear motion, in this case with an ACME profile threaded rod.

In order to make the threaded rod move linearly, its rotation needs to be prevented with something. In this case, it's a sliding mechanism.

Captive in this case means that all the thread, the motor, the sliding mechanism is self contained within the motor assembly, outputting pure linear motion.

An example video can be seen below:

Stepper motors are famous for their silent operation. If you've ever seen a 3d printer in its works you'll see that it makes all kinds of noises. The reason for this is the stepper motors. More specifically the control of the motors. They are not inherently noisy, just they are controlled with a choppy drive signal that causes a lot of harmonics.

One product that helps us combat that is Trinamic SilentStepStick.
Trinamic's StealthChop offers nearly noise-free running of stepper motors.

Linear DC motors:

The linear DC motors work under the same principle as the captive stepper motors. This means they spin an acme threaded rod that's prevented from rotating to achieve linear motion. Often these motors contain a gear reduction, which usually injects some noise into the system. However, they are very cost efficient and easy to work with.

Similarly to BJR_LOG_03, I'll be using a decision matrix to decide which technology to use.

Using the same Measures of performance as before:

- Cost

- Time to prototype

- Visual (or audible) appeal

The evaluation looks like the following:

So moving forward I'll continue the design using Captive linear stepper motors for the actuator.
A quick word on position sensing of these devices.
Steppers can be run open loop without a feedback encoder. The only issue is at startup they have no recollection of where the output actually is. To combat that a homing procedure needs to be executed, where the motor goes towards an end position, and an external switch registers when the output has arrived. For this, I'll be using a micro switch mounted at the bottom of the motor. I'll be experimenting with the stepper controller's StallGuard feature, which...
Read more »