Bloodsport 2 competed on the Discovery Channel’s Season 5 & 6 of Battlebots.
This is our second heavyweight combat robot.

Tech Specs

Weight: 250lbs

Weapons: Overhead Spinning Bar
Weapons Weight: 57-77lbs
Weapons Material: S7 tool steel & AR500
Weapons Speed: 1800RPM (250MPH tip-speed)
Weapons Length: 40-47 inches (1.0-1.2m)

Drive Motors: 4x Short Magmotors
Drive ESC: Brushed Rages
Total Drive power: 14HP

Weapon Motor: 4x Custom Sensored Scorpion 5025 Motors
Weapon ESC: Maytech 200A
Total Weapon Power: 20-30HP

Design & Build Time: 1 year
Location: Boston MA

Team Members
Justin Marple - Captain & Drive Operator
Aaron Lucas - Weapon Operator (S5) & Pit Crew
Nik Buchholz - Weapon Designer & Strategist
Andrew Marple - Weapon Operator (S6) & Pit Crew
Seth Schaffer - Machinist & Pit Crew
Rosa Ruiz - Engineer & Pit Crew
Joran Kiesel - Pit Crew
Matt Marple - Pit Crew
Own Marshall - Pit Crew

History

Following Season 4 Battlebots, we knew the next version of Bloodsport was going to be a serious redesign of the first. After all, the first version wasn’t really meant to be competitive. To flesh out what a new version of Bloodsport might look like, we decided to start designing and building a new Beetleweight that would test out some new strategic ideas and be a kind of test run at a small scale. This turned into a couple of robots that went by a few names: Zero Traction (v3), Mini Laceration, and Paper Cut.

There were several BIG ideas that we knew we needed to answer before designing a new Bloodsport

• What determines the stability of an overhead spinner?

• How can we make an overhead spinner more control oriented?

• How can we make self righting feasible?

Overhead Shenanigans

Probably the biggest and hardest question we needed to answer was, “what on earth makes an overhead spinner stable?”. Tornado Mer and Brutality are two heavyweights that are infamous for having stability issues. Here’s one example: https://youtu.be/WcigMkcgWO0?t=315 . We were quite lucky with the first version of Bloodsport, we never were really plagued by these issues. Though, we knew if we want to improve Bloodsport, stability is something we would need to avoid and understand well.

Since nobody on the team is especially specialized in theoretical physics, we took a more practical approach in starting to understand instability. Our first step was to cut about 10 weapon bars of varying lengths and shapes from AR400 and attached them to our 3lb minibot Thumb war. We then got a series of slow motion videos testing all of these configurations.

Here’s an example:
https://www.youtube.com/watch?v=XI1FDJlE34Y

In general we were able to conclude a couple concepts we thought was fairly obvious: A larger, heavier blade is going to be more unstable. Additionally, a longer blade is also more unstable than a short blade. One discovery that was a bit surprising initially is that foamy squishy wheels are actually worse than stiff wheels. One way to explain this is part of the goal of stability is to try to keep the bar as aligned with the floor we can, and any “wiggle” of the chassis (ie the foam wheels compressing) allows the blade to go unstable easier. Ideally the blade would be bolted to the floor so it would be impossible for the bar to spin out of control, but we only get the drive base in real life, so it’s important the wheels and chassis are very stiff.

Simulations

At this point I think we realized getting any more information from these real-life tests would be too time consuming and costly. So we shifted focus to something that could be scaled easier: simulations! This ended up being a bit tricky because we needed to find a program that could realistically apply the Dzhanibekov effect (or Intermediate Axis Theorem), allow us to script millions of simulations, and give us deterministic/reliable results. We found many physics engines liked to “cheat”, and not take into consideration the moment of inertia of objects, meaning we couldn’t make our bots go unstable.

Without going into too many boring details, we eventually landed on using Gazebo, more famously used for robotics simulations within ROS (robotic operating system). We designed a little testing framework where we could specific weapon bar MOI’s, heights, number of tests, rpms, etc. etc and test various weapon “hits” (usually modeled by a strong upwards force). At this point, we were capable of simulating around a million “hits” per day, when running this on a dedicated aws instance!

Here is when we really started to grow our understanding of the problem. One key correlation we discovered is the “Moment-of-Inertia-ratio”. We found the bar’s stability is largely related to the ratio of the 1st axis to the 3rd axis. For instance, a regular bar that you see on Hazard and Brutality has a quite high ratio, 20:1 or even higher. The lower we can get this ratio, the better off we can be. So that leads to the question: what’s the perfect blade design? Well that would be a 1:1 ratio, which in the ideal world, is a perfect circle. However there is one big issue with a circle, which is that it won’t be very weight efficient. You need to make the circle very thick to make it able to withstand upwards hits without breaking. So what is weight efficient? A tri-bar!

A tri-bar is a great solution for smaller scale robots, as water jetting and material costs are quite low (Zero Traction’s tri-bars cost about $25/ea out of AR500). However when you get to the heavyweight scale, the cost of one is around $2.5k-$3k. This makes it rather impractical to make multiple backup bars. So a cheaper alternative, we can try to manipulate the MOI-ratio of a normal bar by adding “fins”. These fins stretch the 3rd-axis out so a bar with a ratio of say, 30:1, can quite easily become 7:1 or 5:1, which will result in a very stable overhead blade. The obvious tradeoff is that you have to dedicate weight to these fins, which don’t have a huge effect on the energy the blade stores. There are other benefits though, for instance a straight bar is easier to machine, and add pockets too, which will better optimize the bar for vertical hits.

The next consideration for a new bot is control. With the first version of Bloodsport, it had a few key issues: 1) after a hit it took a very long time for the bot to settle back to the ground, 2) when getting in a pushing battle it took very little for the entire bot to be lifted up and have no control, and 3) when we got into a situation where the blade was stopped we had no way to act defensively to get away. The 1st can’t be solved entirely, as often times gyroscopic precision from the blade will keep it from quickly fall onto it’s wheels. Though we can help this quite a bit by moving away from a circular base (see the Black Dragon fight), we can also program the motor controllers to brake to slow the blade down rapidly if needed. The 2nd problem we can fairly easily fix by shifting 2 of the wheels out the back and moving the wheels out as wide as possible. This way, when the bot gets tilted upwards or sideways, we still have some wheels contacting the ground. The 3rd problem can be solved by taking a look at the current vertical spinner meta: wedglets. Wedglets can rest on the ground and get under other bots and wedges, which give an obvious advantage when pushing other bots away. Coincidentally, when you take all of this into account, you end up with a drive base that looks fairly similar to many other 4 wheel vertical spinners. In particular bite force. I can understand why now.

Self righting was the next big challenge to solve. We were able to use on our little beetleweight bots to gain an intuitive understanding of how self righting might work on Bloodsport 2. Through trial and error, we were able to figure out that “curly” pole was an efficient way of getting the bot right-side up. Why does a curly pole work? With a straight “Gigabyte” style pole, there is a huge peak torque requirement on the weapon system at the point the pole touches the ground. By curling the pole, it means the torque requirement is slowly increased as the bot rotates up the pole. It also gives the ability for the bot to “build momentum” as it goes up the pole.

Like before, we found iterating on this design any further will prove to be challenging, so we shifted our focus to simulations. Just like the weapon, we could flip the robot over, run the weapon, and see how much torque it required to flip each design over. And through this, we could gain a better intuition, and even generate, an “optimal” pole design, for our specific chassis and weapon configuration.

So after something like 30000 pole designs and attempting some couple million of self rights, we generated a large spreadsheet which then we could analyze to determine which variables mattered most when self righting (for instance, as we discovered, the pole’s height is very important here, as when the chassis rotates we want to make sure the corners don’t hit the ground).

Season 5 of Battlebots is airing on the Discovery Channel starting December 3rd!

To be continued..