Tuesday, November 2, 2021

So we thought about this a lot!!!

 The challenges for the competition are published here https://piwars.org/2022-competition/challenges/   These are some of our thoughts! I'm sure we will wander off in a completely different direction eventually.

East Devon Pirates PiWars 2022

Competition analysis for robot – ZyderBot

 

The competition has varied in complexity over the years and this year has perhaps provided the most complex challenges.

There are six challenges, two of which are presentation/documentation challenges which we should complete, the blog has already been started and needs regular updates on progress, closer to completion, the technical/artistic merit video must be created.

A competition robot consists of a base chassis to which attachments are fitted to perform the challenges, and the 2022 challenges all require different attachments to be fitted, though it is possible to design a single chassis to perform all challenges, this would not be competitive, not being optimised for any one challenge.

The competition organisers specify chassis dimension limitations, copied here for laziness.



The maximum height is 400mm or 500mm with attachment.

The chassis must carry the main power supply, controller, motive power with ground contact and chassis navigation. It is the main support for all attachments and is used to transport them around the arena. In addition, it must provide connectivity for power and signals to any attachment requiring them as well as a remote control facility.

The chassis must provide a mounting system for attachments which is compatible with the attachments used.

The chassis navigation system must be compatible with the attachments required but will also have to work with additional sensors associated with attachments, such as fruit identification in the picking challenge.

To be competitive in PiWars 2022, the following are requirements.

The chassis must be able to complete each of the arenas challenges autonomously.

The chassis needs to be able to maintain a speed of 500mm/s, including sensor/navigation delays and corners of 200mm radius (reference PiWars 2021 robot performance). Controllable performance in excess of this would be an advantage.

The chassis must be able to position itself accurately to within 12.5mm of a target position. This may be flexible but is supplied as a value by subtracting the maximum width of a robot from the width of the pen door in challenge Shepherds Pi. Typically it must be more accurate than that for competitive speed.

The chassis must carry a sufficient power supply to run the robot for up to 5 minutes with working attachments. This will vary with the challenge, and with challenge 4, the farm tour, the length and intensity of the course.

In previous years, the chassis was the key element being judged, but recently this has changed. Design of the chassis will undoubtably evolve during development of the challenge attachments, and for this reason a prototype with limited performance and primarily remote control usage might be better in the early stages of the project until the attachments are fully functional, perhaps running at ¼ to 1/5 competition speed.

 

There are 4 robot challenges which are detailed on the PiWars website but some details are reproduced here for easy reference.


 

 

Shepherds Pi



Time analysis

The instructions for this challenge only state that competitors should AIM to complete the course three times in 5 minutes. This needs clarification.

The course will be setup originally and will require resetting twice, once between each of the three runs. The course is not complex and the components only have to be placed on the arena at marked positions, the estimate is that each reset will take 30 seconds leaving 4 minutes of run time or 80 seconds per run.

Observations

The distance between wolves is 283mm so should be easily navigated by a robot with the precision noted previously.

The distance between wolf C and the enclosure is 150mm so a full-size robot either cannot negotiate that gap or must move the wolf. Moving a wolf will incur a small competitive time disadvantage but may reduce the time to round up the sheep.

The gate opening is not a requirement but gains additional bonus points. If a gate is fitted, then it must be closed as a separate operation incurring a competitive time penalty. Closing the gate can be achieved easily by pushing it closed, but opening will require an attachment.

A gate opened left to right will pass wolf C with a good margin so can be opened fully, a gate opened right to left cannot open fully unless wolf C is moved. An open gate could be used as a guide to enter the enclosure.

When putting the sheep into the enclosure, the width of the enclosure can only accommodate three sheep before they have to be moved to one side to allow further sheep to be added. An attachment may be required to move the sheep sideways in the enclosure. Moving the sheep will be a necessary time slot.

Making assumptions that suitable attachments are designed and fitted, sheep 1-3 could be rounded up collectively and parked by the edge of the arena. At full competitive speed this might take 10 seconds. The gate is opened using chassis movement and an attachment which might assist. 10 seconds. Sheep 1-3 are collected from the side and moved into the enclosure, taking 10 seconds and then moved to one side, 5 seconds. Total to move sheep 1-3 into enclosure 35 seconds.

Decision to move wolf – no movement

The robot moves through gap wolf B-C and collects sheep 6. 10 seconds, moving forwards collects sheep 5 then 4. 10 seconds, Continuing round wolf A and directly into the enclosure, carefully avoiding wolf C, 15 seconds. Close gate 5 seconds. 40 seconds.

Total 75 seconds

Decision to move wolf – move against enclosure.

Using the sheep attachment, the robot moves the wolf against the enclosure wall, where it is unlikely to fall over. 7 seconds. The robot moves through gap wolf B-C and collects sheep 6. 10 seconds, moving forwards collects sheep 5 then 4. 10 seconds, Continuing round wolf A and directly into the enclosure (at full speed due to no wolf), 10 seconds. Close gate 5 seconds. 42 seconds.

Total 77 seconds

This indicates that there is little time margin between moving a wolf and not, but there is the advantage that moving the wolf reduces the navigation burden.

Both estimate timings are within the 80 seconds per run limit.

The attachment must be able to handle 3 sheep at the same time, but with an attachment plus extension limit of 200mm, 25mm of a sheep will protrude beyond any attachments capacity and must be retained in some way.

The weight of sheep and wolves is not specified, so in construction, weight distribution should put additional weight in the base of each to add stability and the chassis and attachment built to accommodate it.

There are bonus points for opening and closing the gate each time, in these scenarios this would add 10 seconds so is only an option if the run time can be reduced.

Collecting the sheep singly is not an option within these timings.

Collecting the sheep in pairs is not an option within these timings.

Additional bonus points are available for using a whistle control and for recognising wolf and sheep using photographic recognition which is probably essential if not moving wolf C.

Attachments

This challenge is about moving objects around the arena as an obstacle course and the attachments are there to facilitate that.

If a single ‘move’ attachment is to be used, then it must be able to adapt to either sheep or wolf.

In order to move the sheep as quickly as possible, and to facilitate their movement within the enclosure, an attachment must be able to control three sheep at the same time and reach over the top of the enclosure. Imaging must be on the attachment to ensure that sheep are moved reliably, and perhaps use of a tupping marker on the rear of the sheep may help. Navigation by the chassis will need to be located such that it is not obstructed by the attachment and does not inhibit the attachments operation.

An attachment is needed to open and close the gate and this may require a separate vision system or reuse one of the others.

An attachment must also receive whistle commands and relay them to the controller, though this may be a built-in to the chassis.

This is not a stop/go challenge and there needs to be continuous communication between chassis and attachments as to what is happening and when.

Nature’s Bounty



Time analysis

The current rules for this challenge state that it should be carried out three times in five minutes. A competitive run will collect all fruit autonomously, without them touching the arena or being collected by hand. A breakdown of this is as follows.

The tree is set at the start but must be reset twice during the challenge, there are 12 fruit to be accurately located on the tree, with an estimate of 5 seconds per fruit, resetting the tree will take 1 minute. This may be practised and optimised to reduce this time but as a starting position, two minutes of the five minute total will be reset time.

With three runs, there is therefore 1 minute to complete each run.

With a basic tree visiting scenario, in which each quadrant of the tree is visited once, there will be 5 movements by the chassis, estimating 5 seconds per relocation, 25 seconds of the run will be chassis movement time. 

The 35 seconds remaining must be split four ways for each quadrant which (with rounding) will be 9 seconds, and therefore picking time per fruit is 3 seconds. Alternative strategies could be employed to pick two quadrants simultaneously, leaving 15 seconds for movement and 45s second picking, resulting in approximately 4 seconds per fruit.

Using an attachment which picked three fruits simultaneously would give a picking operation available timing of 9 and 12 seconds respectively.

Returning to the barn with a part load of fruit is probably not an option, incurring a competitive time penalty of at least 10-15 seconds.

The chassis and attachment combination must be very stable due to cornering to navigate the tree quickly.

The time for this challenge is very constrained and heavily dependent upon the tree, fruit and attachment design.

Attachments

There are potentially three attachments involved here.

A storage trough to carry all collected fruit must have a capacity of 768ml minimum, and unless the fruit are ordered in the trough in some way, more than that. The bottom of the lowest fruit is at most 65mm from the arena floor, the storage trough therefore has to be below this height, or a separate attachment feature is required to load it with picked fruit.

As noted, the picked fruit must be transported to a temporary storage ‘trough’ while a run is in progress. Higher fruit might be conveyed by gravity, but the lowest fruit either must have the picker carryout the transfer, a separate conveyor mechanism, or the trough placed lower than the lowest fruit, 65mm. One option might be to position the trough below the tree on the front of the chassis and the picker allows the fruit to drop into it. A trough might also have a shaking mechanism to shake the picked fruit into a more space efficient order.

The picker attachment must recognise an individual fruit, grasp/pick a fruit, and transfer it to the trough in the time periods specified. Multiple pickers either as individual operation or operation in concert will either significantly reduce the time required or enable slower picking methods. A multi-picker running at the same speed as a single picker could pick all fruit in 12 seconds, potentially reducing the overall run time to 37 seconds.

The picker must be able to reach the topmost fruit, at 350mm high and be able to pick fruit with a side to side difference of 110 mm, therefore a single picker will need to be able to extend over a wide range and retract to within the 400mm length constraint.  A multi-picker would make the robot at least 350mm high which might present stability issues.

The option to attach via magnets does provide the advantage that reset times can be reduced by making it easier to hang fruit, as well as a clean ‘pick’ action. Fruit using magnets may be attracted to other magnetic materials in the robot or each other, thus causing jams, but also a storage mechanism. The strength of a magnet may also make picking more difficult due to increased force required. If used, magnets may require picking strength tuning.

The chassis navigation will likely be able to position the picking attachments close to the tree and navigate from and to the barn, but more precise positioning close to the tree may need to be done via additional attachment-based sensors.

If the picker or conveyor can retain fruit, the last fruit picked may not require transfer to the storage trough reducing picking time for the final fruit.

The chassis must communicate to the attachment that it is in the correct place to either begin the fine positional adjustments, or to begin picking. The picking attachment must inform the chassis controller that the picking is complete, and the relocation can begin.

Colouring the fruit would significantly aid visual recognition. Can the fruit be distinctively coloured.


 

 

Hungry Cattle

 


This challenge ‘must be run up to three times’ in five minutes. This needs clarification as its not definite?

Time analysis

As before, there are two reset periods when the cattle troughs are emptied, and the robot is refilled. Each trough must be filled with at least 250ml of feed and therefore the robot must contain 750ml of feed. This can be measured in advance but to be certain to adequately fill all troughs to 50%, a margin must be included. However, a reset is unlikely to be longer than 30 seconds if the used troughs are emptied back into the robot attachment or emptied elsewhere and a measured amount used. This leaves 4 minutes for three runs, or 80 seconds per run. There is also the option to prepare the feed by loading enough for three runs before starting, meaning that only the troughs have to be emptied, reducing the reset period to perhaps 20 seconds.

To locate each trough from the start would be around 10 seconds each, or 30 seconds, and 5 seconds to return to the barn, which places the burden of the timing on the fill attachment. A mechanism to dispense feed will operate most quickly with a wide mouth, but also less accurately, and overfilling will use more time. Premeasuring the feed will speed up dispensing it as with the measurement already completed, the feed can effectively be dropped into the troughs.

Feed attachment scenarios.

Holding three runs of feed on the robot would require an initial load of 2.25l of feed, which might be considered an impractical amount, though not impossible.

A single hopper with valve could hold 750-800ml and dispense 250ml based on weight into each trough.

A single funnel could be fed from three premeasured hoppers, dispensing feed via a latch on demand. The end of dispensing would need to be detected

A timed run of 250ml of seed through a 1cm funnel hole took 10 seconds

A timed run of 250ml of seed through a 3cm ‘jar filler’ took 3 seconds

Three separate hoppers could be filled with feed each dispensing to a trough directly.

If using the measured dispenser, dispensing time would be 30seconds in total.

If using multi-hopper with wide funnel would be 9 seconds, but an end of dispensing detector would be required.

If using multi-hopper direct dispensing, dispensing would be 9 seconds, but the navigation would need to reposition to each hopper/trough combination.

It is expected that each trough will be colour coded and that the robot will navigate via a pre-seeded sequence/direction.

These scenarios consider only gravity feed systems, a forced system may be much faster but require significant engineering. I.e. an Archimedes screw.

The run times of 65 seconds and 44 seconds are well within the required times. This challenge may very well become a simple race between teams with the fastest chassis winning.

Attachments

This is the feed hopper and will deliver feed to the troughs. The chassis navigation will need to deliver the hopper to the correct locations and instruct the hopper to deliver the required feed. In turn, the hopper attachment needs to communicate to the chassis that its task is complete. This should be all that is required.

A single hopper is the easiest to position but gives slower feed times.

A multi-hopper might be harder to position but gives faster feed times. 


 

Farmyard Tour

This is the only challenge which relies primarily on the chassis performance only, and even then, only for remote control purposes. One attachment from the Shepherds Pi challenge, the gate opener, might be reused as there is a compulsory element of passing through a removable barrier. However, this challenge relies heavily on the ‘offroad’ ability of the chassis, untested in the arena, and the ability of the operator to control it.

While there is a time limit it is primarily there to limit the time of the run and the whole challenge needs to be designed to fit within the time limit. This will also require inventing a few ’jokes’ to entertain the judges.

A range of attachments remote controlled might also add novelty, such as firing a gun etc.

A trailer can be pulled to offer additional entertainment and possibly contain features, the design of the chassis accommodating the required pulling connection and adding the required pulling power.

No comments:

Post a Comment