Wednesday, August 31, 2011

Piston vs. Gas Turbine Engines, Cost

The small 1.3 HP Honda GX35 engine pictured in the previous post retails for around $250.00.  But no company makes a comparable commercially available, mass-produced, small gas-turbine engine.  The only small 5-50 HP gas turbines that fit this description are the after-market turbochargers made for car engines.  A quick check on eBay shows these items for sale at under $200.00. 

I’m not ready to consider this crude price comparison a valid cost OME (order of magnitude estimate), but it does make me wonder.  It does seem to indicate that, contrary to popular wisdom, a gas turbine might be cheaper to make than a piston engine of comparable power output.

The one area where a valid cost OME comparison might be made is in the area of aircraft engines.  Both piston and small gas turbine engines are used to power private planes and small commercial planes.  So with this in mind, I spent several days searching the web, looking for information on piston versus gas turbines.  It seems the accepted wisdom in the aircraft industry is that, at the small end of the power/size scale, pistons are always less expensive than gas turbines.  But for all of my searching, I never ran across any hard numbers to back up this accepted wisdom.

What comparisons I did find, always struck me as apples and oranges.  Here is the problem: the smaller the turbine, the faster it spins and the bigger the gearbox needed to couple the engine to the propeller.  Why is this so?  A turbine blade’s ability to compress/expand airflow is a function of its speed through that air.  As the radius of a turbine gets smaller, its rotational velocity has to go up in order to keep the blade’s speed constant. 

So as a gas turbine is made smaller it will require an increasingly higher gear reduction to couple to the machinery it is meant to drive.  So it would seem to me, that at some point, it would be the size/complexity of a gas turbine's associated gearbox that drives the cost of a gas turbine engine, not the cost of the gas turbine itself.

Impedance Matching is a term that has specific and quantitative expressions when applied to various problems in physics and engineering.  But it can also be used as a general term to describe the problem of coupling any power source to its load.  For the case of mechanical systems such as an engine driving some mechanism, impedance matching often takes the form of a gearbox. 

For example, if you look at the picture of the Solar Turbines GS-350 shown in the last post, you will notice that the gearbox looks to be about the same size as the turbine itself.  An industrial generator is constrained to work at a 60 Hz frequency; hence the requirement for a gearbox to couple the high speed rotation of the gas turbine to that of its much slower, by comparison, attached generator. 

Gearboxes? We don’t need no stink’n gearboxes!  But unlike industrial situations that require a 60 Hz AC source, our robot harvester/tender is going to be DC powered.  Since our gas turbine’s AC generator’s output current is going to be converted to DC, this means that we can let it turn at any speed it wants too.  Which further means that one can build our small gas turbine generator set as a single shaft connecting turbine rotor, compressor rotor and generator assembly [1].  

And without the cost of a speed reduction gearbox, it appears that the cost of a gas turbine might actually work out to be less than that of a comparable piston engine.

[1] Noting that an added benefit of this configuration is that the generator can be used in reverse as the turbine’s starter-motor.

Sunday, August 28, 2011

Piston vs. Gas Turbine Engines, Power/Size Ratios

When it comes to power/size and power/weight ratios, gas turbines are the undisputed winners over piston engines.  Tanks, heavy lift helicopters and high capacity portable power plants for industry are applications where having the highest possible power/size and/or power/weight ratio is a must.


The M1A1 Abrams uses a Honeywell AGT1500 gas turbine engine [1]
 
Robinson Helicopter Co. R66 gas turbine [2]

Solar Turbines, GS-350, 225 kW generator set [3]
Note: the generator is bigger than the turbine!

The first question that needs to be answered is, what are the size and weight constraints that the design of our robot harvester/tender puts on a possible choice for its power source?   We can get these design constraints by remembering that every design problem always brings with it its own size and weight scales.  So as long as the size and weight of our 'bot's power source is less that 5-10% of its overall size and weight, there is going to be no marginal benefit to making it smaller.

The target size and weight for our 'bot, that we estimated previously, was on the order of 6-8 ft long and less than 300-400 lbs.  This puts a size limit of 1-2 cubic-ft and 10-20 lbs on our 1-2 kW power source.

Here is a 1.3 HP gas engine from Honda that fits our design constraints easily.

Honda GX35, commercial lawn and garden engine [4]

Compared with a small gas turbine from BladonJets that also fits our design constraints. 

Bladon Jets, Micro Gas Turbine Engine [5]
So it seems that either choice of power source, piston or gas turbine, will easily fit into our robot harvester/tender's size and weight design constraints.

And just for comparison, here are a few spec's for a comparable reformed methanol fuel cell.
UltraCell's XX55: 50 W, 3.5 lb, 12.3" 3.2" 8.6" l/w/h

[1] M1A1 AGT1500 spec's
[2] Robinson Helicopter Co. R66's spec's
[3] Solar Turbines
[4] GX35 spec's: 1.0 kW, 7.6 lb, 8" 9.2" 9.4" l/w/h
[5] Bladon Jets

Saturday, August 27, 2011

Piston vs. Gas Turbine Engines

The last few days I've been searching the web, without success, for a nice concise comparison between piston engines and gas turbine engines.  In hindsight, it now appears that my lack of success comes about because both types of engines have their advantages and disadvantages.  And how one engine type compares to the other depends entirely on the particular application.

The main points of comparison between engine types are as follows:

  • power/size and power/weight ratios,

  • thermal efficiency,

  • cost, and

  • reliability and maintenance. 


  • Future Posts: I started this inquiry assuming that a gas turbine would be the clear choice for a small 1-2 kW robotic power plant.  But now I’m not so sure.  In the next few posts I’m going to separately explore each of these areas of comparison with a robot harvester/tender as the target application.

    Tuesday, August 23, 2011

    Agricultural Robotics, BigDog

    BigDog is a gas powered robot produced by Boston Dynamics [1].  This robot design is interesting because it overlaps in many areas with the design of an agricultural robot harvester/tender.  There are some critical differences, though, which, by way of contrast and comparison, can help us better understand some of the design issues facing agricultural robotics.


    Here are some spec's for the BigDog cited on Wikipedia.
    _Dimensions: 2.5 ft tall, 3 ft long
    _Weight: 240 lbs
    _Engine Size: 15 HP go-kart engine
    _Computer: PC/104 stack, Pentium 4, QNX real time operating system [2]
    _Speed: 8 mph
    _Carrying Capacity: 340 lbs 

    Size: One of the main reasons to do order-of-magnitude estimates (OME's) is to get a feel for the size of a design problem before you start.  But doing OME's requires a good engineer's intuition to start with.  The BigDog is a great working example of a robot that fits the profile of a robot harvester/tender.  As such, it offers a great test case to exercise one's OME intuition skills.  

    In physics and engineering, every problem brings with it its own size, mass and response-time scales.  For the case of a robot harvester/tender, it has to be bigger than the plants it will be working on, yet it will have to be light enough to be "wrangled" without the use of extra equipment to pick it up and/or move it around.  This indicates something the size and weight of a motorcycle; that is, dimensions of 3-5 ft and 300-400 lbs or less.

    So, if the BigDog can be used as a good example, then it appears that our attempt at an OME has proven valid. 

    Engine Size: Here is where the design spec's for a robot harvester/tender will differ from those of the BigDog. 

    The BigDog, by the nature of the tasks it's designed to do, will experience high peak loads with large swings between low and high continuous loads.  This calls for a power source that is both "throttle-able" and has a reasonable power response over a wide range of loads.  The engine that fits this requirement is a piston-driven ICE. 

    On the other hand, a harvester/tender ‘bot only needs to move along at a steady speed of 2-20 ft/min.  The tender ‘bot, only needs to carry is its own weight.  While, the harvester ‘bot, will need to carry the extra weight of the produce it’s harvested and the weight of any packing boxes it will need.  Either way, the total work load for a harvester/tender will be lower and much more even than that of the BigDog.  In which case, one can sacrifice the requirements for "throttle-ablity" and wide power-band for a power source with a much greater efficiency than a piston-driven ICE.    

    Next Post: Piston vs. Turbine Engines.

    [1] I would recommend that people check out their website and look at the various robots they build.  Boston Dynamics has its own YouTube Channel , too.
    [2] This is actually not a very large computer core by embedded-systems standards.

    Sunday, August 21, 2011

    Agricultural Robots: Power Source, Batteries, Why Not?

    There are no doubt new battery technologies, being worked on in the lab these days, that will out-perform the Chevy Volt's Li-ion battery.  But for now, the Volt's battery can be considered a good representative of what is "state-of-the-art" for a "light-weight" large capacity battery, manufacturable in production quantities. 

    Chevy Volt Battery, without cover
    It's a "T" shape, roughly 5-ft long and 3-ft wide.  It has a weight of 435-lb (197-kg), and volume of approximately 15.0 cubic-ft.  Its capacity is rated at 16-kWh, but for reasons of longevity, its usable capacity has to be de-rated down to about 12-kWh.

    The Chevy Volt's listed curb weight is 3750-lb, (1750-kg).  The weight of the Volt's battery is not a factor in the car's overall performance because the battery represents only 12% of the car's total weight.  This means that the marginal increase in rolling friction this extra weight brings is only 12%. 

    At 35-mph, the aerodynamic drag for a typical car is comparable to its rolling friction [1].  So the extra drag on the Volt, caused by the extra battery weight, will amount to less than 6% of its total energy consumption when driving down the road at freeway speeds.   This is the reason that the Volt can get away with being battery powered. 

    Now look at the situation for our robot harvester/tender.  Its speed of travel, for most cases, will be on the order of 2-20 ft/min.  Aerodynamic drag will never be a factor.  The overwhelming amount of the energy it uses for locomotion will be expended moving its weight around.  In this case, the weight of the robot will be the dominant factor in determining its power needs.    

    The lighter the robot, the quicker it can move, and the faster it can perform its harvester/tender duties.  Adding a 400 lb battery to a robot that is going to weight less than 300-400 lb to start with, will compromise its performance fatally.

    One could put a smaller battery in our robot harvester/tender, but then it could only run for an hour or so before needing a recharge.  Now imagine trying to run a cluster of ten to twenty robots in a field at one time, each of them needing to be taken out of service every hour or so for recharging.  Our field supervisor would need to bring in extra help to service the 'bots that needed charging, and also bring in extra 'bots to fill in for the 'bots being cycled out of service for recharging.  This scenario becomes a major violation of the WID Rule.       

    So what is the long term outlook for battery powered industrial robots?  One only need look at the specific energies for the different power sources to get an answer. 

    --One gallon of gas, burned efficiently in an ICE, has a specific energy of about 10,000 Wh/kg.
    --The Chevy Volt’s Li-ion battery has a rated specific energy of about 80 Wh/kg [2].

    By weight, a gallon of gas holds over 100 times as much energy as the same weight of a Li-ion battery.  So even if there were some major breakthrough in battery technology that yielded a factor of 10 times the storage capacity over the current Li-ion batteries, it would still fall short of the energy storage capacity of a liquid-fuel based power source by another order of magnitude.

    For a liquid-fuel, energy is stored in the chemical bonds of its molecules; therefore every molecule is an energy storage unit.  In contrast, a battery stores energy in the chemical potential of two ions physically held separate.  Thus, a battery will always require the presence of some inert matrix to hold the two ions apart.  A battery also requires the presence of cathode and anode terminals to provide a pathway for the battery's stored energy to reach the outside world.  In other words, a battery will always contain a lot of extra structure that contributes nothing to its energy storage capacity.  For this reason, on a pound-per-pound basis, a battery will never be able to compete with a liquid-fuel.

    To avoid violating the WID Rule, once a robot is in the field and working, it needs to be able to run without attention for a full 8-10 hr shift.  This requires a power source based on an energy storage capacity that will forever and always be out of the reach of batteries. 

    Sadly for the pro-battery folks, batteries may be a workable solution for robot toys, but they will never be a viable power source for industrial robots that will be required to work in the field for extended periods of time.


    Next Post:  New options for liquid-fuel based power sources

    [1] When I can, I'll get the calculations behind this estimate posted over at my web site.
    [2] For a comparison, a standard 12V lead-acid car battery has a specific energy of about 35 Wh/kg.

    Friday, August 19, 2011

    Agricultural Robots: Power Source

    The choice for a robot harvester/tender's power source will depend on its power requirements.  But what exactly these power requirements are going to be is a big unknown. 

    When, as a engineer, I am faced with a new design challenge, and I don't have any idea were to start, one of the first things I try is to make some order of magnitude estimates (OME's) of the problem I'm dealing with. 

    Coming of age as an engineer, before the introduction of pocket calculators, I did all of my calculations using a slide rule.  A slide rule only works with the significant digits of a calculation; you have to keep track of the exponents separately; so using a slide rule forces you, over time, to acquire the skill of doing quick order-of-magnitude calculations in your head. 

    This is a skill that I don't believe engineering students are being exposed to anymore.  This is unfortunate, since being able to do a quick OME, while not giving you an answer to your design question, does at least give you a good estimate of its "size".  Which, often by itself, will be enough information to let you know which design solutions won't work and help you narrow down the range of design solutions you will need to look at. 

    Another useful aspect of doing OME's is that you can often get a feel for the size of a design challenge, one that you have little prior knowledge of, by comparison with problems you do have past experience with.  I guess one of the observations I'm trying to get at here is that doing OME's requires an engineer to trust his/her intuition on a problem.  But the use of calculators, with their apparent exactness of answers, has taken yet one more opportunity away for students to develop their engineering intuition. 

    So let's try this OME approach to the power source question.  First, I know from my running experience that a runner will burn about 110-140 calories per mile, about 0.15 kWh/mile.  I also recall from memory that a moped got about 100-mpg, about 0.33 kWh/mile.  This suggests to me that a machine will probably consume, maybe, 5-times as much energy [1] as a human to accomplish the same manual labor task.  The other thing I recall from my running experience is that an in-shape human is capable of 100-150 watts of sustained power output over the course of an 8-10 hour workday. 

    Putting these two OME's together indicates that our robot harvester/tender will require a power source on the order of a 1-2 kW's; with a total energy use for an 8-hour workday of about 10-15 kWh's. 

    The next step is to look into possible power sources that fall into this range. 
    • One gallon of gas, burned efficiently in an internal combustion engine, has an energy equivalent of 30-35 kWh; weight, 7 lbs, and volume, 0.15 cubic-ft.
    • The Chevy Volt's Li battery has a usable charge of about 12 kWh, weight, 435 lbs, and volume, approx 15.0 cubic-ft.
    • For comparison, a standard 12V lead-acid car battery holds around 500 Wh.
    • A 1 sq-meter solar panel has a sunny day capacity on the order of 1 kWh per day. But it must be remembered that a harvester/tender robot will be required to work rain or shine, day or night.

    One thing that immediately stands out is the weight and size of a battery capable of powering our robot harvester/tender.  At 400+ lbs, the Li battery will be almost twice as heavy as our robot's frame alone! 

    Future Posts: The only realistic option appears to be a liquid-fuel power source of some kind; this could be either an internal combustion engine or a fuel cell.  Of the ICE’s there have been some interesting developments in the area of micro-turbine engines and, on the fuel cell front, the reformed methanol fuel cell.

    [1] Remember, actual numbers are not critical here, only the general magnitude of the value.  Also, I added some extra design fudge-factor to get to the number 5.

    Monday, August 15, 2011

    The WID Design Rule for Industrial Robotics

    Every time I start thinking about robot design, I keep finding myself running into the same design constraint.  Since I know that I'm going to be invoking this rule over and over again in my coming blog posts, I thought I should articulate it in detail early on.  This way, in future posts, I can just say "The WID Rule".

    The Wild Iris Discovery Design Rule for Industrial Robotics is defined by the following observation.  "It makes no economic sense to replace a worker with a robot if that robot's total cost-of-ownership is not competitive with the total cost-of-labor for the worker it is going to replace."

    I use the phrase "total cost-of-ownership", but I'm not really happy with it.  It's just that I can't think of any better umbrella term that encompasses, in one phrase, the combined cost of everything that using a robot is going to entail.  

    Such as the purchase price factored over the life of the robot, the cost of operation and maintenance, and the labor costs to have someone in the field to monitor the robots as they work.  There are other costs that most people won’t think about, like the cost of the barn or stable facilities where robots will be parked when not in use.  There will have to be some kind of tractor-trailer carrier for moving the robots back and forth from one field to the next, and the labor cost of the driver to operate it.  Then of course there will be registration and license fees and the cost of liability insurance for each of the robots.  OSHA will no doubt get involved with certification and training requirements for anyone going to work around robots.  Anyone who has worked in industry can probably go on and add many more examples of owner/operator type expenses that will go along with the introduction of robots to the farm workforce.  So I’ll just stop here.

    But anyway, back to the discussion at hand.  If I asked any of my fellow engineers about this observation, they would, without a doubt, unanimously say, "yes, of course that's true."  But I've never seen this observation reflected in any of the robot design specs that I've read about.

    Here is an example of how a corollary of the WID Rule gives rise to a design specification.

    WID Rule, Corollary:1, when designing a robot, it is the worker you want to duplicate, not the worker’s task.

    To illustrate what this means, I’ll divide industrial robots into two classes.  The assembly-line factory floor-mounted robot arm characterizes the first class.  In this case, you have a robot intended to replace a stationary worker doing a single repetitive task. 

    Now move from the factory, outdoors to the farm, construction site, logging or mining operation.  The first thing to note is that workers in this second “outdoor” category are not stationary.  As an example from my own younger days as a logger, it was not uncommon at all for me to run/walk five miles or more a day as part of my job.  

    Another aspect of this second “outdoor” category is that tasks are never exactly repetitive; everyday and every work site always brings something new about it.

    And a final aspect of this second category is that workers in these occupations are constantly called upon to do multiple tasks.  As an example, consider the picking of organic leaf lettuce. 

    Since organic fields are smaller, they are not harvested using the harvesting platforms you would see in the larger non-organic fields. In this case, harvesting, cleaning, inspection, packing and loading are all done by the field workers themselves.  Designing a robot that does just one of these tasks doesn’t help the farmer, since he/she will still have to have additional workers or additional specialized robots in the field to do the rest of the remaining jobs.  Thus violating the WID Rule!   

    So to bring this observation back to the question of agricultural robotics, a design engineer needs to look at all [1] of the tasks a field worker is called upon to do during a day’s work and design to that.  To focus on a specific task, such as “picking a strawberry” and then design a robot to do just that one thing might be a very worthwhile educational experience, but it will never give rise to a robot that will find a productive life outside of the engineering lab.   

     [1]  "All" in this case means literally all.  This includes everything from getting off of its carrier by itself, walking out to the field on its own, taking verbal instructions, working autonomously for hours at a time, and walking back to its carrier at the end of its work day.

    Sunday, August 14, 2011

    Dancing Robots vs Robot Dancing

    There must be something characteristic about robot motion that it could inspire distinct hip-hop dance moves. 




    So what is it?  Some might say it is the fluidity of motion, or lack thereof.  Machine motion often strikes us as jerky, abruptly stopping and starting, while animal motion and in particular human motion appears smoother to us

    Is there a technical way to define "fluidity of motion"?  One aspect of fluidity of motion comes from the physics notion of critical dampingThat is, in order to get a mechanical system to execute smooth starts and stops, the accelerations that are applied to the system must be "tuned" to the natural resonance of that system. 

    Often machine motion tends to be characterized by a short startup, followed by constant velocity motion, and ending with a short deceleration to a stop; the quick deceleration often causing the system to wobble.  In contrast, the human body has the natural ability to control its motion to accomplish this effect of critical damping. 

    Another aspect of fluidity comes about from the fact that robots are often programmed to move from one stationarily stable configuration to another via a path which itself is a continuous sequence of stationarily stable configurations. 

    Human movement, in contrast, depends on dynamic stability.  For humans, and animals in general, the positions we go through as we move are not stationarily stable.  That is, if you or I tried to freeze our motion at any one point and hold it, we would fall over; our movement depends on our body’s momentum to carry it through.

    One more thing that characterizes machine motion is that part of the machine will stay fixed while other parts move.  But if you watch a field worker, their whole body will be in motion; that is, their locomotion and picking motions are often one continuous body movement.  And this is the point I’m trying to get at in this post; that for an efficient robot harvester/tender, its locomotion will not be separate from its picking motion. 

    One of the reasons that field workers can do this work, day in and day out, week in and week out, and as quickly as they do, is the smoothness of their hand and body movements.  They don't waste any motions.  The smoother the motion, the less energy it requires, the faster it can go, and the less wear and tear on the system.  And this observation applies to machines just as much as it applies to humans, too.

    I doubt that any robot harvester/tender will be able to go much faster [1] than an experienced picker starting fresh in the morning.  But the advantage that robotics bring to the field is that a robot can sustain that speed all day long without a break.  Speed for a robot harvester/tender isn't the issue.  It's endurance.

    The question then is what methods of robotic locomotion lend themselves to the kind of whole body movements that humans do?  Legs offer the ability for the torso to move as part of the picking/hand/eye motion.  I don't see this being a possibility with wheels or tracks integrated with a simple suspension system.  In those cases, you end up with a rigid torso, which essentially disconnects a robot’s locomotion from its arm/hand/vision-system motion.  I imagine that with a more complicated "active" suspension system, wheels or tracks can be made to mimic the whole-body motion of a legged walker.  But in that case, the robot's locomotion system will end up being just as complicated as a straightfoward walker mechanism.


    Factory floor-mounted robots do have a fluidity of motion.  But they do repetitive tasks, which gives the programmers the opportunity to "dial in” the motion.  For harvester/tenders, the field's surface is constantly changing and the plants are never the same, so there is no opportunity, ahead of time, for a robot harvester/tender’s motion to be programmed for efficiency.

    I found some examples of hexapod-walker motion on the web, but in all cases the motion is very crude.  The only examples I could find, showing a fluidity of motion, were animations such as this of a quad-walker.


    Future Post:  This raises the question is animation a good indicator of the kinds of motion that a robot could potentially be capable of?

    [1]  Regarding the speed of hand movement for a human picker: having watched pick-and-place machines in action doing PCB assembly, I have no doubt that a robot harvester/tender could have hand movements considerably faster than a human’s. The problem isn’t the human hand speed; it’s the plant’s ability to move in response to that hand movement. The stalk, branches, and leaves of a plant form a passive mechanical system with its own natural response time scales. A system can only respond so quickly to outside forces before something breaks or tears. You can try to go faster than a human picker, but I’m afraid, at that point, you risk damaging the plant itself.

    Friday, August 12, 2011

    Agricultural Robots: Locomotion, Walker, Crawler or Wheels?

    Locomotion:  How an agricultural robot is going to move itself around will depend on
    1). the nature of the crop it has to move through,
    2). the specific tasks it has to perform, and
    3). the condition of the field’s surface.

    Tasks such as plowing and harrowing are done using heavy-pull tractors.  Tilling is done using a lighter-duty tractor.  It’s doubtful that the tractor operators responsible for these tasks will ever be replaceable with robotics. 

    The tasks that lend themselves to robotics are those that are currently done by hand, tasks such as thinning, weeding, pre-harvest prep, harvesting, packing and loading.  Also an interesting new option for robotics, that humans currently can’t do, is the task of pest and disease control. 

    Tasks, Harvesters:  Of the hand-harvested crops, you'll see several different field/crop configurations. 

    Strawberries are the only crop that I am aware of that is laid out in wheel-accessible rows that are also maintained that way throughout the planting/growing/harvesting life cycle of the plants.

    Strawberries, Corralitos Road, South Santa Cruz County
    But in general, if a farmer can still see dirt between their plants come harvest time, then they probably aren't getting the yield they could have out of that piece of land.
    Brussels' Sprouts, Jensen Road, North Monterey County
    So, while crops may be planted in rows, by harvest those plants have generally grown to fill the space between the rows.

    For example, some crops, like lettuce or celery, while grown close together, are harvested in one pass.  The harvester platform follows behind the workers picking the crop, so the crop is already up by the time the harvester might roll over the plants.

    Broccoli, Beach Road, Watsonville
    In the photo above, you can see the furrow left by the harvester platform's tires up the center of the image.  Since the part of the broccoli plant that gets harvested is the center, the fact that some of the plant's non-saleable outer stalks/leaves get damaged during the harvesting operation isn't an issue.  A wheeled or tracked harvester would be a workable choice for crops like these.  

    Green Beans, Green Valley Road, South Santa Cruz County
    But there are other crops like green beans or squash that are re-harvested every few days over the two-month (give or take) productive life cycle of the plant.  For crops such as these, "walking" would be the only workable choice for a harvester's method of locomotion.

    Since robot harvesters of this last category will need to be able to "walk" down a crop's rows without damaging the plants they are harvesting, several conclusions follow.

    1). Tethering, that is, dragging a power cord behind, is not going to be a workable design option.

    2). Compromising a walker-robot's agility of movement with a battery pack weighing several hundred pounds will also not be a workable design option.

    3). Computationally, the act of "walking" down a crop's row will consume as much machine vision and motion control resources as any of its harvesting/tending duties.  Not only will these robots have to be able to watch what they are doing, but they are going to have to watch where they are going too!

    Tasks, Tenders:  Activities such as thinning, weeding, pre-harvest prep, packing, loading and pest/disease control fall under this heading. 

    One example of pre-harvest prep is the de-leafing of Brussels’ sprouts prior to harvest.  In this case the field's surface is not only broken earth, but is now covered with a thick carpet of leaves.

    De-Leafing Brussels' Sprouts, location uncertain, North Santa Cruz County?

    A Field’s Surface:  One of the functions of tilling is to get rid of weeds.  Since organic farms cannot use herbicides to get rid of weeds, tilling turns out to be the only effective large-scale way to do this.  But after tilling, a field’s surface will be broken earth, and after irrigation that broken surface turns into soft mud.  This will be the most common surface condition a robot harvester/tender is going to experience.     
     
    Tilled Field, Pioneer Road, South Santa Cruz County


    Next Post:  The motion of an agro-bot's "upper body" is not independent of the motion of its undercarriage.  Undercarriages that have wheels or tracks are going to sway, tip side-to-side and bump as the robot travels down the broken surface of the field it's working in.  This is going to result in a rather shaky/jerky platform for the robot's arm/hand actuators and vision system to have to work from.  One way to deal with this problem is to give a robot an active suspension system of some kind. 

    Another way to deal with this problem is "legs".  Having "legs" gives a robot the means to compensate for the uneven surface it has to travel while also simultaneously being able to adjust the position of its "upper body" to keep its actuator arm/hand/vision platform moving in a controlled fashion.  

    Saturday, August 6, 2011

    Agricultural Robotics, Design Challenges

    Introduction:  Here are the two videos that sparked this blog post.

    The first is a clip of Willow Garage's PR2 in action.  The first thing that struck me was how slow the PR2's movements were.  Then when I went to Willow Garage's web site and checked on the PR2's spec's and price tag of $400K, it struck me how far away from a practical reality the state of the art in robotics still is. 


    The second video was of a demonstration strawberry-harvesting robot in Japan.


    Now compare the speed, dexterity and agility of movement of these two examples with this video of strawberry pickers in the field.


    If you move your attention past the speaker [1] and focus on to the hand movements of the pickers you will get a good introduction as to how quick a robot is going to have to be to compete with existing hand labor.

    Design Challenges:   One of the primary challenges for agricultural robotics is cost.  The cost of labor, to the employer, for a field worker can vary greatly, but for purposes of discussion, let’s just pick a good ballpark value of around $125 a day. Over the course of a 6-7 month season this comes to about $20K. Before a robot harvester/tender can compete with manual labor, its cost-of-ownership must be competitive with the cost-of-labor for the worker it is going to replace.

    There are some crops that lend themselves to machine harvesting; rice and cotton are two examples that come to mind. Farming crops such as these has already been mechanized and automated [2] to the point that robotics offers little advantage to the farmer. But there are far more crops that require hand planting, hand cultivation and hand harvesting for which robotics brings many advantages to the table.


    Cost: the low cost of field labor sets a very low benchmark for the cost of ownership for any robot harvester/tender.

    Quantity: California statewide alone, the number of field workers involved in the hand harvesting of food crops is on the order of 100K, while nationwide, maybe 1M. The potential market for field worker robots is huge.


    Together these two facts indicate that what the robotics industry needs now is a Henry Ford, not another Willow Garage.  Looking at specifics. 

    Power Source: self-powered. For a number of factors, tethering to an external power source is off the table as an option.

    Autonomous Operation: hours to days at a time. There would be no cost savings advantage for a farmer to bring in robotic harvesters if all of the robots required an additional team of high-paid college-educated technicians to shepherd them around.


    Locomotion: walker. Again, for a number of factors, wheels or tracks are off the table as options too.

    Actuators:
    unknown. My personal bet, on the long run, would be hydraulics. But for this to happen will require some kind of new breakthroughs in technology. In the mean time I am staying open to any and all nominations for potential actuator types.

    Operation, Service and Maintenance: K.I.S.S. Since the workforce that will be responsible for these jobs in the future will be the same people that now operate, service and maintain the existing farm equipment; i.e., tractors, harvesters and etc. (*see Autonomous Operation above)

    Machine Vision: the biggie! The computational requirements for machine control are pocket change in comparison to the computational horsepower required to do machine vision. IMHO this is the key element in turning any kind of service robotics from a bench-top toy into a practical reality.


    Future Posts:   In future posts I will be taking up in greater detail, each of the challenges listed above.

    [2].  From the point-of-view of robotics, mechanization vs. automation is a distinction without a difference. My personal usage is to reserve the term mechanization when hand labor is replaced with machines. I use the term automation to refer to the case when a machine is made to run without the benefit of human operators.