Capstone: specifying sensors and actuators for a robot

The previous eleven lessons covered the parts. This capstone puts them together in the way an engineer actually works: starting from what a robot must do, deriving requirements, sizing actuators from loads and duty cycles, choosing sensors to cover the tasks and each other's weaknesses, and checking the result against power, data and budget limits. The example is an autonomous mobile robot (AMR) that carries totes around a warehouse and lifts them onto shelves. The same method applies to an arm joint, a drone or a legged robot, and the worked examples show each step.

Step 1: from tasks to requirements

Good specifications begin with the tasks and the environment, not with parts. The warehouse AMR must carry a payload of up to 60 kg on a 40 kg chassis; drive at up to 1.5 m/s on level concrete with occasional 3 % ramps; accelerate at 0.5 m/s²; stop safely for people; localize to within 2 cm at shelf stations; lift a 20 kg tote 0.6 m in 3 s; run an 8-hour shift on one charge; and work among people, other robots, glass partitions and shiny wrapped pallets. Each statement becomes one or more numbers, and each number has a reason. Writing the reason down matters, because requirements without reasons are impossible to trade off later.

Step 2: drive torque and speed

For a wheeled robot, the wheel torque must overcome rolling resistance, climb grades and accelerate the robot. With two driven wheels of radius r sharing the load, the torque per wheel is:

Step 3: duty cycle and RMS torque

Motors are sized thermally by RMS torque (lesson 8), because heat depends on the square of current. A duty cycle lists the torque in each phase of a typical mission and how long each lasts.

Step 4: choosing motor and gearbox

With torque and speed at the wheel known, choose a ratio and a motor together (lesson 9). A brushless motor with a continuous torque of 0.25 N·m, a peak of 0.75 N·m and a rated speed of 3,000 rpm is a common size. With a planetary gearbox of efficiency 0.90, the ratio must satisfy both speed and torque:

The lift axis is sized the same way. Lifting 20 kg by 0.6 m in 3 s needs a speed of 0.2 m/s and, at constant speed, a mechanical power of \( mgv = 20 \times 9.81 \times 0.2 = 39 \) W. A ball screw with a 10 mm lead must then turn at 1,200 rpm, and the motor torque required at the screw is \( \tau = F\,l/(2\pi\eta) = 196 \times 0.010/(2\pi \times 0.9) = 0.35 \) N·m before acceleration and friction. A spring-applied brake holds the tote when power is lost, because a ball screw is not self-locking.

Step 5: choosing sensors

Sensors are chosen task by task, then checked for gaps. The AMR needs proprioception for control, perception for safety and navigation, and localization at shelf stations.

• Wheel motors: encoders on each motor (motor-side, multiplied by the 15:1 ratio) for speed control and odometry, plus Hall sensors for commutation.
• Lift: an absolute encoder or linear position sensor, and a limit switch as an independent end-of-travel check.
• Body motion: an industrial-grade IMU fused with wheel odometry to bridge slippage and give heading.
• Safety: a safety-rated 2-D lidar at each end, set with protective and warning fields that trigger stops and slow-downs (safety rating and functional safety are covered in ROB-450).
• Navigation and obstacles: a 3-D depth camera facing forward for overhanging shelf edges and objects above the lidar plane, plus a few ultrasonic or optical time-of-flight sensors to catch glass partitions and close-range blind spots.
• Station localization: a downward or forward camera reading fiducial markers at each station, or UWB, to reach ±2 cm where lidar map matching alone is not precise enough.
• Health: motor currents and temperatures, battery voltage, current and temperature.

The gap check asks, for every failure mode in the environment, which sensor still works. Glass: ultrasonic. Shiny stretch wrap that scatters lidar: camera and ultrasonic. A person stepping out from behind a shelf: the safety lidar. Slippery floor: IMU and visual or marker localization correct the wheel odometry. An overhanging fork tine: the depth camera. Where the answer is 'nothing', the sensor set is incomplete.

Step 6: power, data and budget checks

Finally, check the totals. Sensors and computers draw power continuously, motors draw it in bursts, and the battery must cover the whole shift. Data rates must fit the robot's networks and computer. And the bill of materials must fit the budget, which often forces a second iteration.

import math
# Data budget for the AMR's perception sensors
sensors = {
    "2 x safety lidar (2-D)":   2 * 1100 * 25 * 8,          # points/scan * Hz * bytes
    "depth camera 640x480":     640 * 480 * 30 * 2,         # 16-bit depth at 30 Hz
    "RGB camera 1280x720":      1280 * 720 * 30 * 3 // 20,  # roughly 20:1 compressed
    "marker camera 640x480":    640 * 480 * 15,             # 8-bit mono at 15 Hz
    "IMU + encoders + ultrasonic": 200 * 64 + 4 * 1000 * 8 + 8 * 20 * 8,
}
total = 0
for k, bps in sensors.items():
    total += bps
    print(f"{k:28s} {bps/1e6:7.2f} MB/s")
print(f"{'total':28s} {total/1e6:7.2f} MB/s  = {8*total/1e6:6.1f} Mbit/s")

2 x safety lidar (2-D)          0.44 MB/s
depth camera 640x480           18.43 MB/s
RGB camera 1280x720             4.15 MB/s
marker camera 640x480           4.61 MB/s
IMU + encoders + ultrasonic     0.05 MB/s
total                          27.67 MB/s  =  221.4 Mbit/s

The depth camera dominates the data budget, as it usually does. Around 28 MB/s, about 220 Mbit/s, is comfortable for a USB 3 or gigabit Ethernet link and a modern embedded computer, but logging it all for an 8-hour shift would need roughly 800 GB, so the robot will record selectively. Checks like these, done early, prevent the unpleasant discovery late in a project that the chosen sensors cannot all be connected, powered or processed together.

In the United States, an AMR like this one would also be designed to the safety standards that apply to industrial mobile robots, such as ANSI/A3 R15.08 and ANSI/A3 R15.06, and operated under OSHA's general duty to provide a safe workplace. Its radio modules would carry FCC authorization and its lidars a valid laser classification under FDA rules. ROB-450 covers these requirements.