teaching_robots_navigation

Evaluation

Example of LIDAR

LiDAR works like “laser radar”—it fires invisible light beams and measures how long they take to bounce back to calculate precise distances. Here is how the LiDAR was evaluated in this ground robot report, along with how that same process translates to flying drones (UAVs).

Ground Robot LiDAR Evaluation (This Project)

LiDAR Evaluation for UAVs (Drones)

In aerial robotics, LiDAR isn’t just looking at barrels—it is mapping terrain, avoiding power lines, or holding altitude. Evaluating a UAV’s LiDAR focuses on four different real-world challenges:

More resources

Teaching Module: Sensor Evaluation & Engineering Trade-Offs in Robotics and UAVs

Course: Advanced Mobile Robotics / Unmanned Aerial Systems (UAVs)

Target Audience: Undergraduate / Master’s Level Engineering Students

Topic: Empirical Sensor Evaluation, Failure Modes, Power/Weight Trade-offs, and System Integration

1. Executive Summary & Pedagogical Goals

When designing autonomous systems, students often assume sensors operate under ideal mathematical assumptions. In practice, sensor selection is driven by physical environment noise, payload weight limits, and battery constraints.

Key Learning Objectives

  1. Analyze physical sources of sensor inaccuracy across LiDAR, cameras, and IMUs—with special emphasis on UAV dynamics.
  2. Evaluate the power-weight-throughput triangle in aerial vs. ground robotics.
  3. Formulate concrete sensor fusion strategies that compensate for individual sensor failure modes without exceeding onboard energy budgets.

2. Sensor Inaccuracies & Failure Modes

A. LiDAR (Light Detection and Ranging)

LiDAR measures time-of-flight (![][image1]) of emitted laser pulses to compute distance ![][image2].

B. Visual Sensors & RGB-D Cameras

C. Inertial Measurement Units (IMU) & GPS/GNSS

3. Power, Weight, and Energy Constraints

In ground robotics, adding a ![][image14] sensor suite increases motor draw slightly. In aerial robotics (UAVs), payload mass penalizes flight time non-linearly.

The UAV Payload Energy Penalty

Hover thrust required is directly proportional to total mass ![][image15]:

![][image16]Power required for a multirotor to hover scales according to momentum theory:

![][image17]Adding a ![][image18] mechanical LiDAR package to a ![][image19] drone increases mass by ![][image20], but increases hover power draw by approximately:

![][image21]+———————————————————————–+
| THE SYSTEM TRADEOFF |
| |
| +——————-+ Higher Weight +—————+ |
| | Heavy Sensor | -————————> | Short Flight | |
| | (High Precision) | | Duration | |
| +——————-+ +—————+ |
| | ^ |
| | High Data Rate | |
| v | |
| +——————-+ High Wattage +—————+ |
| | Onboard Compute | -————————> | Battery Drain | |
| | (GPU/Edge AI) | | | |
| +——————-+ +—————+ |
+———————————————————————–+

Sensor Category Comparison Matrix

Sensor Type Typical Weight Power Draw Accuracy Range Primary Failure Mode Best Engineering Use Case
Mechanical LiDAR ![][image22] ![][image23] High (![][image24]) Vibration, High Power, Mass High-precision 3D Terrain Mapping
Solid-State LiDAR ![][image25] ![][image26] Medium-High (![][image27]) Limited Field of View (![][image28]) UAV Frontal Obstacle Avoidance
Stereo Camera ![][image29] ![][image30] Medium (Degrades with range) Low light, Featureless walls Close-range VIO & Landing Alignment
Ultrasonic Rangefinder ![][image31] ![][image32] Low (![][image33]) Soft surfaces, Angular limits Low-altitude Ground Proximity Hold

4. Classroom Case Studies & Practical Exercises

Case Study 1: The Multi-Robot Contaminated Barrel Cleanup (Ground)

Case Study 2: Powerline Inspection UAV (Aerial)

5. Homework & Lab Assignment Ideas

Assignment 1: Sensor Trade-Off Calculations

Task: Given a ![][image40] baseline quadcopter with a ![][image41] battery (![][image42]), calculate total hover time under two payloads:

  1. Option A: Stereo Camera Suite (![][image43], ![][image44] power draw).
  2. Option B: Mechanical 3D LiDAR (![][image45], ![][image46] power draw).

Students must apply the ![][image47] relationship and account for sensor electrical draw to plot flight time degradation.

Assignment 2: Failure Mode Analysis & FSM Design

Task: Design a state machine (or RoboChart) for an inspection drone encountering sensor degradation.

6. Seminar Discussion Prompts

  1. “Why might a $50 ultra-sonic sensor be a better choice for low-altitude UAV altitude hold than a $1,500 lightweight LiDAR?”
    (Key points: Power draw, mass penalty, low compute needs, reliability over flat non-absorbing surfaces).
  2. “If a system’s reported sensor accuracy is high in simulation, why does performance often degrade significantly when deployed on real hardware?”
    (Key points: Sensor noise models in simulators are often Gaussian and ignore environmental interference like prop-wash, glare, temperature drift, and dynamic vibration).
+——————-+ Higher Weight +—————+

| | Heavy Sensor | ————————-> | Short Flight | | | | (High Precision) | | Duration | | | +——————-+ +—————+ | | | ^ | | | High Data Rate | | | v | | | +——————-+ High Wattage +—————+ | | | Onboard Compute | ————————-> | Battery Drain | | | | (GPU/Edge AI) | | | | | +——————-+ +—————+ | +———————————————————————–+ Sensor Category Comparison MatrixSensor TypeTypical WeightPower DrawAccuracy RangePrimary Failure ModeBest Engineering Use CaseMechanical LiDAR$500\text{g} - 1.5\text{kg}\(10\text{W} - 35\text{W}$High ($\pm 1-3\text{ cm}$)Vibration, High Power, MassHigh-precision 3D Terrain MappingSolid-State LiDAR$100\text{g} - 300\text{g}\)2\text{W} - 6\text{W}$Medium-High ($\pm 3-5\text{ cm}$)Limited Field of View ($60^\circ-90^\circ$)UAV Frontal Obstacle AvoidanceStereo Camera$50\text{g} - 150\text{g}\(1.5\text{W} - 3\text{W}$Medium (Degrades with range)Low light, Featureless wallsClose-range VIO & Landing AlignmentUltrasonic Rangefinder$10\text{g} - 30\text{g}\)<0.5\text{W}$Low ($\pm 5-10\text{ cm}$)Soft surfaces, Angular limitsLow-altitude Ground Proximity Hold4. Classroom Case Studies & Practical ExercisesCase Study 1: The Multi-Robot Contaminated Barrel Cleanup (Ground)Context: A ground robot must approach, attach to, and transport toxic barrels.Failure Encountered: The 1° LiDAR beam misses small target barrels at range $> 3\text{ m}$, returning inf.Engineering Outcome: Re-architect system roles. Use the camera (blob detection) for mid-range tracking and estimation, restricting LiDAR to close-range ($< 1\text{ m}$) collision prevention.Case Study 2: Powerline Inspection UAV (Aerial)Context: Autonomous drone scanning thin powerlines ($1-2\text{ cm}$ diameter) at $15\text{ m}$ distance.Failure Encountered: Solid-state LiDAR FOV misses thin wires when the drone yaws rapidly. Stereo cameras fail due to sun glare.Engineering Outcome: Combine a high-frame-rate solid-state LiDAR ($100\text{ Hz}$) with an IMU-driven kalman filter. Enforce a flight speed cap ($v \le 2\text{ m/s}$) to bound braking distance within the maximum detection range.5. Homework & Lab Assignment IdeasAssignment 1: Sensor Trade-Off CalculationsTask: Given a $1.8\text{ kg}$ baseline quadcopter with a $4\text{S } 5000\text{ mAh}$ battery ($74\text{ Wh}$), calculate total hover time under two payloads:Option A: Stereo Camera Suite ($120\text{ g}$, $2.5\text{ W}$ power draw).Option B: Mechanical 3D LiDAR ($850\text{ g}$, $18\text{ W}$ power draw).Students must apply the $P \propto m^{1.5}$ relationship and account for sensor electrical draw to plot flight time degradation.Assignment 2: Failure Mode Analysis & FSM DesignTask: Design a state machine (or RoboChart) for an inspection drone encountering sensor degradation.Condition 1: Camera lens is covered in dust (Image entropy drops below threshold).Condition 2: LiDAR returns zero valid points for $> 500\text{ ms}$.Required Student Solution: System must safely transition to a stable altitude hold using IMU/barometer and initiate a controlled vertical land or return-to-home protocol.6. Seminar Discussion Prompts”Why might a $50 ultra-sonic sensor be a better choice for low-altitude UAV altitude hold than a $1,500 lightweight LiDAR?”(Key points: Power draw, mass penalty, low compute needs, reliability over flat non-absorbing surfaces).”If a system’s reported sensor accuracy is high in simulation, why does performance often degrade significantly when deployed on real hardware?”(Key points: Sensor noise models in simulators are often Gaussian and ignore environmental interference like prop-wash, glare, temperature drift, and dynamic vibration).