Team 1678 received 8 falcons the week of thanksgiving and began testing the Saturday (30 November 2019). Thus far, our tests have included using a 6 Falcon WCD, 4 Falcon WCD, 1 Falcon wrist (rotating position-controlled mechanism), and 1 falcon rolling cargo intake. All these tests were performed on our 2019 practice robot.
Drivetrain
Gearing remained the same as our miniCIM drive for convenience (7.2:1, 4in wheels). For 4 Falcon WCD tests, the third falcon on each side was unplugged and allowed to spin freely.
Our drivetrain testing involved both closed-loop and open-loop performance.
Open-loop Tests
Open-loop performance was measured with a sprint test. We used a 4 Falcon WCD setup for our sprint tests, as 6 Falcons caused severe carpet slippage due to excessive torque at the chosen ratio. Our results:
Sprint to 250in, voltage manually ramped to 12V over 0.1s: 1.78s
Closed-loop Tests
A variety of spline auto paths were run to determine closed-loop performance. Particularly important to us was consistency in terms of real-world measured “error” with auto paths that had high accelerations. Driving consistency was key in games like the 2018 PowerUp where consistent high precision (without reference of a vision target in most teams’ case) was required to intake cubes throughout nearly the entire 15 second autonomous period. That consistency comes from an effective control loop, dependent both on the internal encoder on the Falcon and the linearity on the motor for predictable characterization.
Drive path following was more consistent than our MiniCIM drive at similar acceleration/velocity/voltage constraints. Paths were followed using a Ramsete controller, and a simple kV, kT, kS style feedforward model characterized to an r^2 north of 0.998.
We were pleasantly surprised at the real-world repeatability of drivetrain trajectory following. The consistency was apparent with a high curvature spline going 60 inches forward and 30 inches to the right, which ended within 0.25in of real-world error. Video of the spline:
(Equivalent Accuracy was achieved with a 4 and 6 Falcon setup with wheel slip as the only limiting factor for acceleration constraints)
Wrist Tests
The wrist gearbox was modified to provide a 72:1 reduction, as compared to the old 254:1 reduction when it had been used with a 775pro. Both wrists were controlled using the MotionMagic control mode provided by CTRE (a trapezoidal motion profile). We got the desired accuracy on the Falcon wrist with a P gain and theoretically calculated F gain.
A video comparison of the two wrists where both control loops tuned to peak in the 9-12V range (falcon on the right):
When the motion profile was tuned to aim for a similar speed and acceleration (making the falcon peak at 6-ish V) as the 775pro wrist, the falcon proved far more efficient.
A power input comparison (W vs time (s) and deg vs time (s)):
Nearly the same position vs time for both wrists (in the red box), but the Falcon one is a bit faster with more load (carrying a piece of cargo). This means the Falcon wrist is following a harder energy path. Yet, the falcon needs about 25% the peak power input.
Cargo Intake
The cargo intake with Falcon 500 was geared to 3:1 reduction via a versa planetary. The system formerly ran a 775pro with a 10:1 reduction. We saw no noticeable performance improvement between systems.
Overall Impressions
-
Drivetrain performance using the internal encoders for odometry in combination with CTRE’s pigeonIMU was more than sufficient, even for high curvature + high linear acceleration applications.
-
High power meant snappy instant torque at lower speeds; very nice for quick accelerations in position-controlled mechanisms (see wrist video comparison).
-
The high stall torque makes it feel quick while driving (able to accelerate to 10 fps in just 0.2 seconds in sprint test).
Feel free to ask any further questions regarding our results beta-testing these motors.
Logs shown:
Disclaimer: 1678 has been a member of Team IFI since 2016.