← Projects FTC Robotics · INTO THE DEEP · 2024–25

Into the Deep FTC Robot

Team Captain  ·  FTC Team 15083 Overclock

Reaches into the submersible to pick up samples, scores them in the high basket or hangs specimens on the high chamber, then climbs to a level 3 ascent, using Kalman-filtered localization, Pedro Pathing autos and OpenCV sample detection.

Top 2%offensive rating (OPR)
27–9–1season record
3judged awards
843D-printed parts
CAD render of the robot, front three-quarter view
Cypher · 15083 Full robot

Hover over a subsystem

Full robot 01Drivetrain 02Elevator 03Laterator 04Intake 05Climber 06Localization
Sample match
Mechanical · 01

Drivetrain

Compact mecanum base with an underside sweeper.

Drive base with mecanum wheels, motors and the underside sweeper
Drive base
Underside of the drive base showing three dead-wheel odometry pods
Underside: three low-mounted dead-wheel pods
  • Mecanum drive to strafe around the submersible, which takes up most of the field.
  • 12:1 drive gearing.
  • Servo-driven underside sweeper clears samples from under the submersible bar in auto.
Mechanical · 02

Elevator

Stacked vertical slides that reach the high basket and high chamber.

Robot with the elevator at the selected height
  • Stacked Misumi SAR240 slides on each side, extending to over 4½ ft.
  • String-driven continuous lift.
  • Carries the laterator, intake and specimen claw, which closes automatically when a color sensor sees a specimen.
  • Encoder positions for single-button scoring (section 07.4).
Reverse-image specimen claw with an optical sensor and a flat alignment wall
Specimen claw
Mechanical · 03

Laterator

Horizontal slides mounted on the elevator, so the robot works like a crane.

Robot with the laterator at the selected extension
  • Reaches about halfway into the submersible to grab samples, then retracts.
  • Riding on the elevator means no transfer between separate mechanisms.
  • Can reach over obstacles, and over an alliance partner to score.
Mechanical · 04

Intake

Pop-out arm with a 270° wrist and an active grip-wheel intake.

Intake arm and wrist at the selected position
  • Pop-out arm driven through a gear train that takes load off the servo spline.
  • 270° wrist to pick up samples in any orientation.
  • Compliant grip wheels powered by a double-helical gear train and a goBILDA servo: “touch it, own it.”
  • A limit switch stows the arm automatically after a pickup.
  1. Pyramid claw
    • Couldn't grab samples sitting next to each other
  2. Inverse claw
    • Weak grip; couldn't pick up specimens
  3. Active intakeFinal
    • Picks up from any direction with little driver precision
Mechanical · 05

Climber

3D-printed pivoting slides for a level 3 ascent.

Climber at the selected rotation and extension
  1. Grab
    • Hooks the low rung with the slides at an angle
  2. Rotate
    • Slides pivot and lock vertical to shift the center of mass
  3. Extend and pull
    • One-way latches hold the low rung while the slides reach the high rung, then the robot pulls itself up
Robot on the rungs with the one-way latch marked
One-way latch on the low rung
3D-printed climber slides and their rotational pivot
3D-printed slides: 10× cheaper than aluminum
Mechanical · 06

Localization

Three dead-wheel odometry pods plus a camera for AprilTags.

  • Dead wheels track the robot's real movement, so wheel slip and collisions don't throw off its position.
  • A camera reads AprilTags around the field to correct odometry drift.
  • A Kalman filter fuses the two measurements, weighting each by how much error it carries, for a position more accurate than either alone (section 07.2).
Software · 07

Software

Averaged about 13 samples in the high basket or 10 specimens on the high chamber per match, with a top 2% OPR and one of the highest scores in Texas.

07.1

Autonomous

  • Pedro Pathing: paths drafted, simulated and timed in Pedro's path generator.
  • 4-sample high-basket and 5-specimen high-chamber autos, with time left over.
  • Several programs so we fit any alliance partner's auto.
  • Paths are built from Hermite splines through waypoints, giving smooth, continuous motion between scoring positions.
  • Sensors close the loop: odometry corrects the path in real time, and the color sensor and limit switch confirm each pickup before the robot moves on.
Field diagram with 4-sample and 5-specimen autonomous paths
4-sample and 5-specimen autos
Cubic Hermite spline between waypoints $$H(t) = (2t^3 - 3t^2 + 1)\,p_0 + (-2t^3 + 3t^2)\,p_1 + (t^3 - 2t^2 + t)\,m_0 + (t^3 - t^2)\,m_1$$
\(p_0,\ p_1\)
start and end waypoints
\(m_0,\ m_1\)
tangents (heading and speed) at each waypoint
\(t\)
progress along the segment, 0 to 1
07.2

Localization

Odometry drifts about 3% and depends on a known start; AprilTags are accurate but not always visible. A Kalman filter weights each source by its error, leaning on AprilTags when they're close or available, which takes odometry error from 3% to nearly zero.

Kalman update $$\hat{x}_k = \hat{x}_{k|k-1} + K_k\left(z_k - H\hat{x}_{k|k-1}\right)$$
\(\hat{x}_{k|k-1}\)
prior estimate (odometry)
\(z_k\)
AprilTag measurement
\(K_k\)
gain: how much to trust the tag
07.3

Sample detection

  • OpenCV pipeline (in development during the season) finds dense patches of yellow, red or blue pixels in the webcam feed.
  • Draws bounding boxes and picks the nearest sample in the submersible.
  • The robot collects it to score, or hands it to the human player to become a specimen.
OpenCV detections drawn on colored samples
OpenCV sample detection
07.4

Driver automation

  • Single-button scoring: encoder setpoints for the slides and climber hooks, combined with on-the-fly path generation and auto-drive.
  • Localization runs in teleop too, enabling zone-specific settings like lower speed near the baskets.
  • The specimen claw's color sensor closes the claw and raises the elevator automatically, saving about 2 s per cycle.
  • The climber detects current spikes to prime its hooks consistently.
  • Field-centric driving from the IMU, so the driver can rotate and translate at once.
07.5

State machines

Each mechanism's behavior is a finite state machine: defined states (driving, intaking, scoring) with clear transitions on inputs and sensor events, instead of long chains of if-statements. That makes autonomous routines more predictable and reliable.

Season

Iterations

  1. Meets 1–313–4–1
    • Level 2 hang arms and a claw intake
    • 1st, 2nd and 2nd place
    Robot at Meets 1 to 3
  2. Regional Qualifier5–1–0
    • Geared arm and new 3D-printed climbing slides
    • Finalist alliance captain; Innovate Award
    Robot at the Regional Qualifier
  3. League Tournament5–1–0
    • Two-wheel intake, wider slide spacing, more consistent hang
    • 4th in qualifications; Think Award
    Robot at the League Tournament
  4. Area Championship4–3–0
    • Added the laterator (“extendo”) for longer reach into the submersible
    • Playoffs; Design Award
    Robot at the Area Championship