RBE 2002: Unified Robotics II: Sensing

01 Background

RBE 2002: Unified Robotics II^[1] was the counterpart to RBE 2001. Where 2001 focused on actuation — converting electrical power into controlled mechanical motion — 2002 focused on the inverse problem: how a robot's control and decision-making are informed by its sensors. The course covered the operation and integration of simple and complex sensors, including signal transduction, interface circuitry, and physical integration, as well as techniques for handling uncertainty through signal conditioning, digital filtering, calibration, parameter selection, and sensor fusion. Recommended background was ECE 2049: Embedded Computing in Engineering Design^[2].

Labs applied classroom knowledge including operational amplifier design, oscilloscope measurement, strain gauges for bend and force sensing, voltage dividers, debounce RC networks for mechanical switches, potentiometer-tuned resistor banks, and sensor-driven control logic running on an Arduino microcontroller.

02 Signal Conditioning & Sensor Fundamentals

Early lab work exercised the building blocks of sensor signal conditioning. We tested op-amp circuits on the oscilloscope, varying the feedback resistor networks to hit specific closed-loop gain targets and observing how signal shape, offset, and noise floor changed with component choice. Strain gauges — both off-the-shelf and laser-cut test articles we fabricated ourselves — translated mechanical deflection into an analog voltage we could capture on the scope, giving us a direct handle on how much bend energy was going into a material.

We also built debounce circuits for mechanical switches from scratch. A push-button wired directly to a GPIO pin registers dozens of false transitions as its contacts physically bounce on closure; the fix is a small RC network sized to filter out the bounce transients while still responding crisply to a real press. We dialed in resistances with potentiometers on a breadboard before committing to fixed values on the final robot.

03 Final Project: Dry-Erase Board Cleaning Robot

The final project was a contained, judged competition: a dry-erase whiteboard covered in scribbles, mounted vertically. Each team's robot hung from an overhead carriage that the teaching assistant drove at a fixed rate across and up-and-down the board — teams did not control the gross motion. What we engineered was the eraser end of the carriage: how it deployed, how it retracted, how hard it pressed, and the pattern it traced. Scoring was visual — after one controlled pass, whichever team left the least marker residue won.

The constraint that shaped every decision was the frame. Every team was given an identical size envelope by the TA, and every mechanism had to fit inside it. Inside that envelope, weight distribution turned out to matter as much as the envelope itself: too much mass behind the eraser and the robot tipped back, lifting the eraser off the board; too much mass forward and the robot sagged, breaking contact on the return sweep. The pivot-and-moment-arm calculations we had just covered in lab turned into a physical balancing act — we had to analytically predict the center of gravity and then confirm it against the scale.

There was one more wrinkle in the physics. The carriage was hoisted from a motor mounted at the top of the board structure by a single cable, which made the robot, the cable, and the board face a right triangle: the cable was the hypotenuse, the board face was the long leg, and the robot hanging off it was the short leg. As the carriage travelled up or down the board, the cable angle relative to the board changed — steep near the top of the board, shallow near the bottom. That changing angle decomposed the cable tension into different fractions of normal-into-the-board and tangent-along-it, so the eraser's engagement pressure was not constant: it varied with the robot's vertical position. Calibrating for a single, static normal force wasn't going to cut it.

04 Chassis & Eraser Deployment

The frame itself was built from VEX Robotics angle iron and structural metal pieces — an erector-set-style component system that was quick to assemble, easy to reconfigure, and rigid enough to handle the moment-arm loads without flex. Using VEX stock for the frame let us spend our design time on the mechanism that actually mattered — the eraser deployment — rather than fabricating a one-off chassis from scratch.

The deployment itself was a rack-and-pinion linear slide. A pinion gear driven by a motor pushed a linear rack forward, translating the eraser into the board on engagement and retracting it cleanly between passes. Running deployment on a dedicated linear track — rather than hinging the eraser on a rotating arm — meant the contact pressure was dictated by the slide's end-of-travel and the robot's static weight, not by a servo's stall torque. Predictable contact force made the rest of the calibration tractable.

The eraser-mounting bracket was the one custom part. I dimensioned it in AutoCAD against digital-caliper measurements of the physical pieces it would carry — the eraser body, the VEX mounting holes, the link-arm geometry — and printed it on my Prusa i2 at home. Once a drawing was locked I could print a bracket overnight, test-fit the next morning, and iterate. The final print held the eraser firmly, was light enough not to upset the moment balance, and was cheap enough in plastic to print spares in case one cracked under competition load.

05 Four-Bar Linkage & Erasing Pattern

Our first eraser pattern ran the eraser through a circular motion in place, sweeping its contact patch around a fixed point on the board. In practice this produced uneven pressure around the circle — high-pressure arcs on one side, low-pressure arcs on the other. The high-pressure zones burnished marker into the board instead of lifting it; the low-pressure zones left streaks. Coverage was mixed.

The redesign mounted the eraser on a four-bar linkage built from a couple of VEX extension arms, driven off the deployment slide. As the linkage cycled, the eraser's contact patch traced a figure-8 rather than a static circle — two overlapping loops that migrated chaotically across the board during each engagement. The irregular motion redistributed pressure more evenly and broke up the burnish pattern — noticeably better erasing than the rotating version, at the cost of a couple of extra metal arms and pin joints. It was a good practical lesson in how a small change in kinematics can move a design from “mixed” to “works.”

06 Debouncing, Force Sensing & Calibration

With the mechanism built, integration was about translating sensor readings into control decisions. Push-button contact switches, mounted to detect when the eraser had fully engaged the board, had to be debounced with the RC circuits we had developed in lab. A single physical contact without debouncing produced a burst of GPIO transitions that the Arduino interpreted as rapid re-engagement — the motor would chatter and the slide would hunt around the contact point instead of settling.

Force sensing let us measure the normal force between the eraser and the board in real time. Knowing this closed the loop on the weight-balance analysis: given the carriage's pivot location and the robot's measured center of gravity, we could predict what the eraser normal force should be and then verify the sensor agreed. The supporting math involved deriving calibration coefficients that mapped raw ADC counts into pounds-force, which we then cross-checked against a physical scale during weigh-in.

The weigh-in served double duty: confirming the robot was within the TA's competition mass envelope, and anchoring the moment-arm math. Too heavy and the normal force exceeded what the slide's return mechanism could retract against; too light and there wasn't enough contact pressure to lift marker off the board. Legal mass and a centered center-of-gravity was the sweet spot — and you could see the robot's footprint on the glass partitions in the ECE building we tested against, a visible record of exactly how much board area the eraser was actually contacting per pass.

07 References