TBBT: Triathlabots

The fourth episode of The Big Brain Theory asked teams to develop robots that could compete in three athletic events; a 100-meter dash, a javelin toss, and a standing long jump. The dash and the javelin toss were won by the team that went the fastest and furthest, but the javelin toss was governed by who tossed the javelin closest to the Olympic record. You can catch the full episode here.

MY BLUEPRINT CHALLENGE

I started building robots in freshman year of high school by participating in the US FIRST competition, a yearly competition where teams get a big kit of parts and 6 weeks to build a 130-pound robot. This particular athletic event felt a lot like building a FIRST robot; those robots are typically required to do a number of difficult tasks, and combine those tasks into one chassis. My immediate answer to this challenge was to start thinking about how I built those robots, and how I build most robots these days; let’s build it from scratch!

Episode 4 Blueprint

I provided a loosely sketched concept to the judges (with a fixed javelin position fired by stored energy, a fixed high-pressure pneumatic piston to launch the robot, and gigantic outrunner brushless motors to power the drivetrain), and focused on the most important qualities of that robot; that it have an extremely high power-to-weight ratio for all of its subsystems, that it push through its center of mass with the jumping piston so it didn’t spin out of control in mid-air, and that it have as few moving parts as possible to increase system robustness.

None of this was particularly clever; it was simply an attempt to design the most robust, highest-performance robot possible. In my opinion, it lost out because it wasn’t all that innovative, it was just sound engineering; this develops into a running theme throughout the series.

One quick note – it wasn’t clear from the way the challenge was described if we were supposed to imitate humans or simply do similar types of tasks. That’s why you see Eric and Amy proposing legged vehicles that were generally not going to be competitive against wheeled vehicles; it wasn’t clear which way the judges were going to decide. What was also not clear (and announced later) was that the jumping element of the competition allowed for a secondary robot that was 25% of the mass of the full vehicle to detach from the main system, in order to jump on its own. These are both more instances in which we didn’t feel like we had all the information we needed from the start to come up with good designs, leading to more frustration.

TEAM PICKS

I was glad to be picked first for a team, but I was incredibly anxious about working with Dan as a team leader. Most of Challenge 3 had been fairly calm, but towards the end he had exploded in the same way he had blown up in every previous challenge, making me wary of working with him in general. That said, I hadn’t seen him in a leadership role, so I was willing to give him the benefit of the doubt going into it. I was excited to be working with Amy (another US FIRST veteran), as we had started seeing eye to eye over the course of the last challenge and shared a lot of fabrication and robot-building experience.

DESIGN

Our brainstorming phase was nowhere near as cohesive as they made it seem in this episode. Let’s review the major subsystems, and talk through all of the different questions we had to address.

Episode 4 Ideation

We approached the ideation process in a way similar to how Amy tended to handle the start of a challenge – by drawing everyone’s ideas on the board. I had had a bit more time to think about a design after feeling like the judges weren’t interested in the fully custom system, so I proposed simplifying the design process by buying a vehicle we could modify – a giant-scale remote control car. We agreed on this fairly quickly, and I immediately went to a hobby store to purchase an electric 1/5th Scale HPI Baja (in addition to all the RC electronics we would need to create the rest of the robot), a 4-wheel-drive monster of an RC truck that weighed 20 pounds and could hit speeds as high as 60 miles per hour.

We then proceeded to spend a lot of time discussing strategies for firing the javelin. The javelin in question weighed two pounds, and we had to launch it as close to the Olympic world record distance as possible (which I believe was around 300 feet). We immediately ran into a problem when we realized that the size of the RC car severely constrained the potential size of the javelin launcher, which meant that we had to use a very energy-dense launching system. We thought about an air cannon (similar to Blue Team’s design) briefly, but discarded it due to weight concerns. The two major ideas boiled down to a compound bow suggested by Joel and a latex tubing-based launcher that I suggested.

Compound Bow Launcher

I was initially enthusiastic about the idea of converting a standard bow into a launcher, because it seemed very well tuned for what we wanted to do. Upon calling an archery store and speaking to a world-class archer and bow designer, however, we were informed over the phone that such a launcher would self destruct within just a few shots. The largest compound bow available was designed to fire no more than 150-grain arrows (which themselves are considered heavy, and generally only designed for crossbows), which weigh approximately .021 pounds. We were trying to fire a 14,000-grain javelin. Heedless of the advice of professionals and several team members, Dan insisted that we spend $1,000, hours of our time to pick it up, and yet more hours of our time to test it just to make sure. By comparison, two latex tubes that we used to compare launchers shot the javelin 30 feet farther on their own in our test runs. The compound bow eventually self-destructed due to test fires before the competition day, making the question of launcher type a moot point.

Episode 4 Javelin Launcher

The final question came down to how to jump. During the Blueprint Challenge, we all believed the entire robot had to jump. The judges later clarified that a smaller, jumping robot that weighed 25% of the mass of the larger robot, could detach and jump on its own. We came up with three major ideas; jump the entire RC car with a high-pressure pneumatic system, jump a smaller car with a spring-based system, or jump a smaller car by detonating an airbag (suggested by Joel). Everyone, including myself, loved the airbag idea – unfortunately, it was shot down by the Powers That Be because it was technically an explosive. This left us to decide between a high-pressure pneumatic system that would jump the large car, and jumping a small car with a spring system.

Jumping Car

Dan and I had discussed some of the robots I had worked on and seen at Boston Dynamics, that had similar capabilities to what we were attempting to do. As a result, he wanted me to create a high-pressure pneumatic system that would allow us to jump the whole robot, and told me as much at the outset of the competition. Unfortunately, we soon hit a major snag – this particular competition started on a Friday, and ended on a Monday. As a result, every single industrial supplier I used on a regular basis (excluding standard industrial suppliers like McMaster and Grainger, which don’t carry the parts I was looking for) was closed for overnight shipping to California before we could even come up with a design. As soon as I realized this, I told Dan that I couldn’t source the parts in time, and we needed to switch designs. He insisted that I keep pushing on the pneumatic design regardless of how many times I let him know I couldn’t get the parts. Eventually, after two days of searching and calling any store I could find that was open on a Saturday (including Lowlife Hydraulics, a shop that specialized in making lowrider cars jump up to 10 feet in the air with hydraulic pistons!), I cobbled together a design that mixed paintball air storage tanks, single-acting hydraulic cylinders, Chinese knockoff dump valves, and hydraulic fittings and hoses to create an 800 psi pneumatic jumping system. I’m not even going to post a systems diagram here – it was a terrifyingly dangerous mixture of components that only went together out of desperation and a lack of the proper suppliers. It didn’t work too well (it lacked an accumulator, and relied on the flow-limited outlet port of the paintball tank to supply air), and wouldn’t integrate into the existing design anyway. By the time Dan was finally ready to scrap the idea and switch to a spring-loaded jumping robot, we only had one day left in the build. To this day, Dan believes I somehow screwed him over for not being able to deliver a high-pressure pneumatic system out of thin air.

One final note – I designed, ordered, implemented, and tested the entire electrical system on this vehicle. Both the large robot and small robot were controlled by a single remote controller, the entire electrical system (including 3 independent battery systems, 4 motor controllers, 2 receivers, 3 hobby servos and a variety of other components) was built on the last day of the competition in under 4 hours, and it all worked perfectly the first time. None of that is mentioned, of course.

WASTES OF TIME

2WD Dyno

See that device in the scene with the judges standing around the table? That’s a two wheel drive dynamometer. While I was picking up the RC car, Dan had Joel spend the better part of 3-4 hours building this contraption. I honestly don’t know why it was built in the first place – traditional dynamometers stress test a motor’s ability to produce torque, and give you information about how fast a motor (or car, in this case) can go. In our case, we had no way to change the torque setting of the dynamometer, we had no sensors on the dynamometer, and on top of all that, we had a four wheel drive car – which meant that the front wheels spun quickly when the back wheels were loaded by this device. At no point was it useful for anything. This challenge was fraught with this kind of pointless time-sink. Notable wastes of time included building the 2WD dynamometer for a 4WD car (3-4 hours), working on a high-pressure pneumatic system after I identified I couldn’t source parts (8-10 hours), purchasing, installing and testing the compound bow after being told it couldn’t work (8-10 hours), and a variety of smaller tasks after that.

At one point, Amy was responsible for designing and machining parts for a screw system that would be used to draw the latex bands back for firing the javelin. It would replace the winch we had been using, and would save us a few pounds (in exchange for 6-8 hours of labor). At the start of the last day, Amy identified that she couldn’t make the parts fast enough to complete the device. Dan instructed her to work on the device anyway, ‘because what else would you do for the last 6 hours of the build’? (It’s worth noting that at this point in time, the electrical system didn’t even exist yet.) It took the entire rest of the team stating in no uncertain terms that we needed to install the winch instead and move on with the design for Dan to back down, and even so he remained unhappy about it.

ME AND AMY

This episode didn’t show the camaraderie that Amy and I developed over the course of this challenge, which might make later episodes more confusing. We both worked to our full capacity, learning to work together in both design and manufacturing, and we developed significant respect each others’ skills.

PERFORMANCE

As expected, we completely annihilated the dash event. We beat Usain Bolt’s record-setting pace by 2.5 seconds. In terms of power-to-weight, there are few vehicles on this earth that beat a remote control car in a short straightaway. Then, it was time for the javelin toss.

Javelin Firing

This was one of the biggest shames of this show. There’s no reason that we shouldn’t have been able to design a latex spring slingshot launcher that shot a javelin 300 feet, especially one that was drawn back by a 2,000 pound capacity winch. In the end, we suffered from a failure of not having sufficient strength in a hobby servomotor to pull back on a common, off-the-shelf release mechanism. If we had tested at full load, we likely would have identified that failure mode and fixed it. We would also be able to see that we didn’t have enough latex bands on the launcher to hit the distance required. As it was, we never tested at full javelin pull because we simply didn’t have the time, and as a result were blindsided by that failure in the competition. And then, of course, there was the “jumping car”.

Real Jumping Car

I’m not sure there’s anything I can say about that.

POST MORTEM

In hindsight, the choice of an RC car base (even one as large as the 1/5th scale car we found) severely limited our design choices by forcing us to build small – the act of designing small, robust assemblies is incredibly difficult, and building parts for those small assemblies is even harder. It would’ve certainly been possible to create a much more robust javelin launcher that weighed about the same as our final system, but even so, it would’ve been hit and miss to attempt to hit the target distance of 300 feet repeatedly (because remember, the challenge was to come as close to 300 feet as possible – not go as far as possible). As far as the jumping car is concerned – if we had spent three or four days designing that system well, we should have been able to come up with something competitive. We only had to launch 5-10 pounds, compared to their 60 pound mini-car. It would have been a difficult design process to get right, as it involved a lot of balancing and coming up with a robust rail solution, but it would’ve been possible. Had we spent our time very well, I would have given us much better odds (let’s say, 50/50) of beating Blue Team. As it was, we stood no chance.

My hat is off to the Blue Team in this episode. They had fantastic team dynamics, and those dynamics resulted in a great robot. They planned their strategy well, and dominated the events they chose to focus on. The pneumatic piston they chose to use featured a standard, low-pressure system (though it was heavily modified), which meant they had access to parts from standard industrial suppliers. Deploying that system meant that their “mini-car” weighed 20 pounds more than our entire robot.