Miniature Rotorcraft as Aerial Explorers
Ilan Kroo, Peter Kunz
Dept. of Aero/Astro
Stanford University
NASA/DoD Second Biomorphic Explorers Workshop
JPL Dec. 5, 2000
Outline
Introduction
Challenges
Development approach
Cooperative control issues
Summary and future directions
Introduction
Objectives:
Examine feasibility of small autonomous rotorcraft
Explore scaling issues and limits on feasible size
Develop some of the required technologies
Bio-Inspired aspects:
Insect-scale aerodynamics
Testbed for cooperative control / swarm behavior
The Concept: Meso-scale Flight
What is a meso-scale vehicle?
Larger than microscopic, smaller than conventional devices
Mesicopter is a cm-scale rotorcraft
Exploits favorable scaling
Unique applications with many low cost devices
Objectives
Is such a vehicle possible?
Develop design, fabrication methods
Improve understanding of flight at this scale
The Concept: Rotorcraft
Why rotorcraft for meso-scale flight?
As Reynolds number and lift/drag decrease, direct lift
becomes more efficient
Compact form factor, station-keeping options
More flexible take-off / landing
Direct 4-axis control
Scaling laws (and nature) suggest cm-scale flying devices
possible.
The Concept: Applications
Atmospheric Studies
Windshear, turbulence monitors
Biological/chemical hazard detection
Planetary Explorers
Swarms of low-mass mobile robots for
unique data on Mars, Titan
Terrain-independent
Aerial Explorers Complement Rovers
Planetary Explorer Missions
Accompany rovers
Atmospheric sampling
Imaging / mapping
Search
Earth, Mars, Titan
Features of Small Rotorcraft
Rotorcraft
Low ground speed
Operates in restricted areas
No runway requirement
Inefficient?
Small Vehicles
Favorable structural scaling
Lower cost (especially transport)
Many small > few large
The Concept: Challenges
Insect-Scale
Aerodynamics
3D Micro-Manufacturing
Power / Control /
Sensors
Challenges: Aerodynamics
Insect-scale aerodynamics
Highly viscous flow
All-laminar
Low L/D
New design tools required
Approach
Advanced aerodynamic analysis and design
methods
Novel manufacturing approaches
Teaming with industry for power and control
concepts
Stepwise approach using functional scale
model tests
Approach: Aerodynamics
Navier-Stokes analysis of rotor
sections at unprecedented low
Reynolds number
Novel results of interest to Mars
airplane program
Nonlinear rotor analysis and
optimization code
Aerodynamics: Section Optimization
Nonlinear optimization coupled with
Navier-Stokes simulation
New very low Re airfoil designs
Improved performance compared with
previous designs
Section Optimization
Preliminary solution
bears strong resemblance
to dragonfly section
(Newman 1977)
Structural advantages to
insect section
Optimized Solution
Aerodynamics: Section Flight Testing
Micro sailplanes permit testing of
section properties
Difficulties with very low force
measurements in wind tunnel
avoided
Optical tracking system
Aerodynamics: Rotor Optimization
Chord, twist, RPM, blade number designed
using nonlinear optimization
3D analysis based on Navier-Stokes section
data
Rotor matched with measured motor
performance
Approach: Rotor Manufacturing
1. Micro-machine bottom
surface of rotor on wax
3. Remove excess
epoxy
2. Cast epoxy
4. Machine top
surface of rotor
5. Melt wax
Rotor Manufacturing: Materials and Methods
Wide range of rotor designs
fabricated and tested
Scales from .75 cm to 20 cm
Materials include epoxy,
polyurethanes, carbon
Power and Control Systems:
Sensors / Control Laws
Innovative passive stabilization
under test at larger scale
Linear stability analysis suggests
configuration features
MEMS-based gyros provide
damping
Approach: Prototypes
Initial 3g device with external
power, controllers
Basic aero testing complete
Issues: S&C, electronics
miniaturization, power
Approach: Prototypes
Capacitor powered mesicopter
5mm Smoovy
Integrated electronics
Shrouded frame
Approach: Prototypes
Low cost unaugmented 60g
system
Includes receiver, speed
controllers, lithium batteries
Closed loop control using offboard vision
Approach: Prototypes
PC-board system with
digital communication and
on-board microcontroller
Mesicopter Development: Prototypes
Prototypes
From 13g to 200g
Flight video
Mars Rotor Development
Very low Re
environment
Tests in Mars
atmosphere
simulator at
JPL
Mesicopters as Cooperative
Control Testbeds
Ideal for studying collaborative control strategies (CO, COIN)
Multi-resolution mapping mission
Decentralized control and navigation
DoD / NASA applications
Real, 3D problem features
Control Approaches
Self-organizing systems display interesting emergent
behavior.
Self-optimizing systems display desired emergent
behavior.
Approaches here employ nonlinear optimization, exploit
recent progress in distributed design and large-scale MDO.
Focus on high-level control, planning
Control Approaches
Centralized design
Behavior of each agent determined by system-level control law
Heuristic rules
Individual actions determined by global rules, local data
Reduced basis optimization
Rules used to reduce dimensionality of optimal design problem
Distributed design
Individuals seek local goals leading to desired system properties
An Example Application
Simple example to illustrate approaches:
Formation flight of geese
Goal is not just to maintain formation, but to optimize
performance
Include aerodynamic interactions, test control concepts
Formation Flight of Geese
Each bird leaves wake that influences
others. Drag includes viscous, selfinduced, interference.
Objective to is maximize the range of
the group (minimize drag of least
fortunate individual).
Control is individual speed
Consider coplanar formation (optimal)
Centralized Design
Nonlinear optimization used
directly to find best speed and
position.
Works in steady case for
limited size flock, good initial
distributions.
Fails completely in other
cases, scales poorly.
Heuristic Rules
Assume V-Formation
Set Vi = V0 + k (xi xi-1 Dx0)
Specify reasonable values of V0, k, Dx0
Drag reduction is achieved
Requires little communication
Reduced Basis Rule Design
Use rule to reduce design dimensionality
Optimize V0, k, Dx0 using nonlinear programming for
steady state solution or Monte Carlo.
This works but:
Not robust. Individual parameters sensitive to uncertainties, disturbances
Not correct (sub-optimal)
Distributed Design
Concept:
Let each individual seek best local solution.
Choose objective definition and decomposition to produce system
optimum.
Exploit previous work:
Collaborative optimization and MDO
Collective Intelligence concepts
Distributed Design
Collaborative optimization (CO):
Multi-agent control problem analogous to large scale multidisciplinary
design optimization problem.
CO is a multi-level decomposition and design strategy developed to solve
this.
Collective intelligence (COIN):
Ideas under development in AI (NASA Ames) to help select local
objectives.
Useful in traffic management, economics, network routing.
Distributed Design Example
Greedy objective: every goose flies at speed that
minimizes his or her drag with interference
Result: Tragedy of the Commons
Example
Distributed Design Improved
Basic idea:
Modify local objectives to include effect on others.
Specific idea:
Vote on best speed to fly, then fly at Vi = ( k1 Vv + k2 Vi*)
k2/k1 determines self-interest or altruism
Distributed Design Improved
Control History
Collaborative, rulebased control law
Distributed Design
Result is a robust
method that efficiently
produces correct
solutions with limited
communications
Distributed design
approaches allow think
globally, act locally to
work.
Mesicopter Status
5 self-powered prototypes at various scales
Largest (200g) can carry video, INS, digital
FCS and fly for 15 min
Successful closed-loop hover demo using offboard vision
Work continues on FCS, simulation, optical
flow stabilization, inter-vehicle
communication
Future Work
Near-term applications
Testbed for multi-agent, cooperative control
Earth-based tests
Longer term aspects
Mars rotorcraft
Alternate power source potential
Further miniaturization of electronics
Acknowledgements
Work supported by:
NASA Institute for Advanced Concepts
JPL
Langley, Ames
Work undertaken by:
Profs. Prinz, Kroo
5 Stanford Ph.D. students
