1 of 28

Critical Design Review

FLYING HAND

——Aerial Manipulator for Non Destructive Testing

TEAM C

Advisors

Dimitrios Apostolopoulos

John Dolan

2 of 28

Team Members

Shubham Saboo

Shunxuan Wang

Chenyang Mao

Shubha Tanaya

Weerachai Sangiumpaisankij

2

3 of 28

01

System Overview

4 of 28

Workflow

4

Workflow Diagram

5 of 28

Overall System

5

6 of 28

02

Demonstration

7 of 28

Force Application

8 of 28

Subsystem Core Components

20

9 of 28

Subsystem Workflow

21

10 of 28

Subsystem Capabilities

22

The force sensor can measure contact force and calculated desired thrust output from the Wrench Controller

11 of 28

Contact Demo

24

12 of 28

Achieved Requirements

23

ID	Functional Requirement	ID	Performance Metric
MF1	The system shall sense the surrounding environment	MP1	Sense the environment within the range of 0.5-4 m from the Flying Hand
MF2	The system shall localize its own pose relative to the environment	MP2	Localize position with an error of at most 10 cm under nominal conditions
MF3	The system shall allow operator to manually pilot Flying Hand to an Area of Interest	MP3	Navigate at speeds up to 1 m/s during manual flight
MF5	The system shall manipulate end-effector to establish contact with the POI	MP4	Detect a POI from a distance of at least 2.5 m
MF6	The system shall maintain stable end-effector contact during NDT	MP5	Maintain end-effector positioning accuracy within ±0.15 m during contact-based NDT
MF10	The system shall provide real-time visual feedback to the operator

13 of 28

Localization

14 of 28

Subsystem Core Components

14

Voltage Regulator

Jetson Orin Nano

D456 RealSense Camera

Flight Controller Unit

USB-TTL-232-Serial

HDMI EDID Emulator

D456 RealSense Camera

Jetson Orin Nano

Flight Controller Unit

Voltage Regulator

USB-TTL-232-Serial

HDMI EDID Emulator

15 of 28

Subsystem Workflow

15

Mavros�/mavros/odometry/out

16 of 28

Subsystem Capabilities

16

Open-Area

Do waypoints navigation and

land automatically within 0.5 m of target point

Near-Wall

~0.3 m Away from the Wall

~1.5 m Flight Height

Hover for 30 s

Arm-End Keyframes During Near-Wall Hovering

17 of 28

Near-Wall Demo

18

18 of 28

Mission Video

18

19 of 28

Validating Wavepoint Navigation Performance

18

EKF Local and Global Estimates

Flight trajectory as seen on the Map

20 of 28

Benchmarking VIO Performance with Ground Truth using RTK-GNSS Module

18

RTK segment — length: 24.85 m | angle: -48.9°

VIO segment — length: 25.11 m | angle: 165.5° | straightness RMS: 0.270 m

── Comparison ─────────────────

Length error: +1.0% (VIO 25.11m vs RTK 24.85m)

Heading error: -145.6° (VIO 165.5° vs RTK -48.9°)

VIO straightness (RMS from line): 0.270 m

"This slide covers how we validated VIO performance against an independent ground truth. Since the drone had no onboard GPS, we mounted a separate flight controller with an RTK-GNSS module on the same airframe, logging position independently. The RTK log captured one straight side of the box trajectory — roughly 25 metres. We then geometrically matched that segment to the corresponding straight run in the VIO trajectory, without needing any timestamp synchronization between the two systems. The result: VIO reported 25.1m against RTK's 24.85m — a length error of just +1%. Lateral straightness was 27cm RMS over that run. The 145° heading offset you see in the overlay is not a flight error — it's a one-time frame initialization difference between the two flight controllers. Bottom line: on this segment, VIO position accuracy is within 1% of RTK ground truth, which gives us confidence in the scale and linearity of the estimate."

21 of 28

Achieved Requirements

17

ID	Functional Requirement	ID	Performance Metric
MF1	The system shall sense the surrounding environment	MP1	Sense the environment within the range of 0.5-4 m from the Flying Hand
MF2	The system shall localize its own pose relative to the environment	MP2	Localize position with an error of at most 10 cm under nominal conditions
MF3	The system shall allow operator to manually pilot Flying Hand to an Area of Interest	MP3	Navigate at speeds up to 1 m/s during manual flight
MF5	The system shall manipulate end-effector to establish contact with the POI	MP4	Detect a POI from a distance of at least 2.5 m
MF6	The system shall maintain stable end-effector contact during NDT	MP5	Maintain end-effector positioning accuracy within ±0.15 m during contact-based NDT
MF10	The system shall provide real-time visual feedback to the operator

22 of 28

Flying Hand

Simulation

23 of 28

Subsystem Core Components

26

Fully-actuated

F/T sensor

Camera

Target Localization

24 of 28

Subsystem Workflow

27

Mission Script:

The Axis Control:

What Affects X, Y, Z Motion

The Blending: How the Three Loops Combine

Slide 3 — Subsystem Workflow

"Now let's understand the simulation workflow. The backbone of this subsystem is three nested control loops. First, a Position Controller — essentially a PID on position error — which holds the drone at fixed waypoints. Second, Image-Based Visual Servoing, which detects the point of interest using the RGB camera and commands velocity directly from pixel error. And third, the Wrench Controller — a PID on force error from the desired contact force. The simulation runs a mission script: takeoff, hover to a fixed location, approach the point of interest, make contact, hold the desired force, then retreat — all performed autonomously. Different controllers own different axes at different phases. To be explicit: until the drone reaches the hover point, all axes are controlled by the Position Controller. Once there is visual lock on the target, Visual Servoing takes control of the X axis — driving the drone forward toward the target — and the Y and Z axes for lateral alignment. As we approach contact, through to the moment the desired force is achieved, Visual Servoing and the Wrench Controller share authority over X through the hybrid controller. So — let's see this control blending live in simulation."

25 of 28

Simulation Demo

30

26 of 28

Subsystem Capabilities

28

Effective Distance of Visual Servo Controller > 2.5 m

Arm-end Contact Error < 0.15 m

Force-convergence Time after Contact < 3 s

Maintain a Constant Force 5±1 N

Duration of Force Application ≥ 15 s

27 of 28

Achieved Requirements

29

ID	Functional Requirement	ID	Performance Metric
MF1	The system shall sense the surrounding environment	MP1	Sense the environment within the range of 0.5-4 m from the Flying Hand
MF2	The system shall localize its own pose relative to the environment	MP2	Localize position with an error of at most 10 cm under nominal conditions
MF3	The system shall allow operator to manually pilot Flying Hand to an Area of Interest	MP3	Navigate at speeds up to 1 m/s during manual flight
MF5	The system shall manipulate end-effector to establish contact with the POI	MP4	Detect a POI from a distance of at least 2.5 m
MF6	The system shall maintain stable end-effector contact during NDT	MP5	Maintain end-effector positioning accuracy within ±0.15 m during contact-based NDT
MF10	The system shall provide real-time visual feedback to the operator

28 of 28

Thank You!

TEAM C