Flapping-wing flight in bird-sized UAVs for the Robur project: from - - PowerPoint PPT Presentation

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Flapping-wing flight in bird-sized UAVs for the Robur project: from an evolutionary

• ptimization to a real flapping-wing mechanism
• E. de Margerie

J.-B. Mouret

• S. Doncieux

J.-A. Meyer

• T. Ravasi

P . Martinelli

• C. Grand

ISIR-Université Paris 6 CRIC (IUT Cachan)

MAV 2007

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Birds and UAVs

Birds are better than current UAVs

extremely maneuverable (perching, slow flight, sharp turns) energetically efficient (gliding, fast forward flight)

Part of these capabilities originate from complex wing kinematics ➔ closed-loop control of wings ➔ no-sinusoidal kinematics ➔ many degrees of freedom

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Robur project

Robur project : design and control a bird-sized flapping-wing UAV, from the point of view of bio-inspired artificial intelligence

neural-network controllers evolutionary algorithms bio-inspired behaviors (e.g. soaring, optic flow, ...)

Bird-sized (versus insect-sized) :

Soaring is possible High-payload (artificial intelligence onboard) Outdoor flight

Institute of Intelligent Systems and Robotics (ISIR, Univ. Paris 6) and IUT Cachan (CRIC)

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Robur artificial bird

Basic features/choices : position-controlled wing-beat mechanism ➔ arbitrary movements rigid panel-based wings (easier to simulate and to build) articulated wings (wing folding and twisting) closed-loop control ➔ different from most current designs (toys, slow-hawk, ...) In this talk : wing-beat mechanism

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Typical experiment

Goal : Closed loop control of forward flight Tethered flight on a whirling arm Aerodynamic measurements Learning experiments (evolution of neural network controllers) No free flight ➔ no weight constraints

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Topic

Problem : we want to explore complex flapping-wing kinematics with this experiment but ... How to choose the right motors for the wing-beat mechanism ? allow the right angular ranges ?...

flapping frequency ? power ? angular ranges ? wing-span ?

➔ basic kinematics are required to design a wing-beat mechanism ➔ a mechanism is required to test efficient kinematics ➔ “chicken-and-egg” problem

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Approach

• 1. Evolutionary optimization in simulation

simple kinematics parameters morphologies (wingspan, aspect ratio, ...)

➔ typical flight speed, mechanical power, angle ranges, ... ➔ specifications

• 2. Mechanical design

➔ classical engineering

• 3. (future work) whirling arm experiments

➔ evolution of neuro-controllers

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Approach

• 1. Evolutionary optimization in simulation

simple kinematics parameters morphologies (wingspan, aspect ratio, ...)

➔ typical flight speed, mechanical power, angle ranges, ... ➔ specifications

• 2. Mechanical design

➔ classical engineering

• 3. (future work) whirling arm experiments

➔ evolution of neuro-controllers

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Approach

• 1. Evolutionary optimization in simulation

simple kinematics parameters morphologies (wingspan, aspect ratio, ...)

➔ typical flight speed, mechanical power, angle ranges, ... ➔ specifications

• 2. Mechanical design

➔ classical engineering

• 3. (future work) whirling arm experiments

➔ evolution of neuro-controllers

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Approach

• 1. Evolutionary optimization in simulation

simple kinematics parameters morphologies (wingspan, aspect ratio, ...)

➔ typical flight speed, mechanical power, angle ranges, ... ➔ specifications

• 2. Mechanical design

➔ classical engineering

• 3. (future work) whirling arm experiments

➔ evolution of neuro-controllers

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Evolutionary optimization

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Simulated UAV

0.5 kg 2 rigid panels by wing 4 degrees of freedom (DOFs) by wing : dihedral, shoulder twist, wrist twist, wing folding (sweep) simulator : semi-empirical (validation : polars and wind tunnel) airfoil : Selig 4083 control : sinusoidal curves

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Parameters

Optimized parameters wing area (0.1-0.4 m2) wing aspect ratio (4.5-10) flapping frequency (1-10 Hz) amplitude of rotation for each DOF

• ffset for each DOF

time offset with the dihedral ➔ 12 real parameters Ranges chosen according to zoological data corresponding to birds of similar masses.

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Evolutionary algorithm

Fitness, two objectives to be optimized simultaneously :

flying along the most horizontal path given a target speed mechanical power used (to be minimized)

Multi-objective evolutionary algorithm (ε-MOEA, Deb 2005) This algorithm try to find the set of all

• ptimal trade-offs between objectives at

the same time (Pareto-optimal) ➔ no weight between objectives 24 evolution runs, for target horizontal speed ranging from 6 to 20 m/s

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Results : videos

6 m/s 12 m/s

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Results : power

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Results : wing folding

drawing by Karl Herzog

wing folding was used for all flying speeds medium speed : 25-44% of power decrease high speed : 7-17%

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Results : useful data

Typical (most efficient) flying speed : 10-12 m/s Minimum power consumption : 20 W/kg Medium to high speed : 20-60 W/kg Wing folding decreases substantially power consumption Typical flapping frequencies : 3-5 Hz Angles : Speed (m/s) Dihedral Should incid. Wrist incid. 6-8 15-50 0-30 10-50 10-12 25-45 0-15 8-15 16-20 30-65 0-5 1-10 chosen ± 50 ± 30

• ➔ Specifications for a real flapping mechanism (dihedral and

twist)

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Mechanical design

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Overview

To left wing To right wing Dihedral parallel mechanism Shoulder incidence parallel mechanism Conical gears Pulley-belt components

Patent pending Minimum energetic consumption for a sinusoidal movement Other kinematics are possible Two rod-crank parallel mechanisms Symmetrical movements

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Overview

To left wing To right wing Dihedral parallel mechanism Shoulder incidence parallel mechanism Conical gears Pulley-belt components

Patent pending 4 brushless motors (30 W, 100g) 0-5 Hz Dihedral ± 50 deg. Twist ± 30 deg.

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Kinematic schema

a b L L J1 J2 J3 J4 J5 J6

Wing Wing

ϑ ϑ

Motor 2 Motor 1

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Simplified schema

a b L L λ ϑ ϑ A3 A4 A1 A2 A5 A6 α1 α2 u1 u2 a b γ

a b L L λ ϑ ϑ γ α1 α2 Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Variables

a b L L λ ϑ ϑ γ α1 α2

New input variables α (mean input angle) and ϕ (half-phase angle). α = 1

2(α2 + α1)

ϕ = 1

2(α2 − α1)

and α1 = α − ϕ α2 = α + ϕ ϑ = f(ϕ, α) = sin−1 L − λ 2a λ =

• L2 − 4b2 cos2 α sin2 ϕ + 2b sin α sin ϕ

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Quasi-sinusoidal motion

• 0.3
• 0.2
• 0.1

0.1 0.2 0.3 1 2 3 4 5 6

Evolution of the flapping angle for different phase angles ϕ α (rad) θ (rad)

ϕ=.25 rad ϕ=.50 rad ϕ=.75 rad

Motors at constant speed ➔ minimum energy consumption ➔ quasi-sinusoidal motion ˙ α = 2π · fϑ ϕ = sin−1( a

b sin(ϑmax))

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Pseudo-periodical motion

To obtain a pseudo-periodical motion : ➔ modification of the quasi-sinusoidal motion The motor velocities have to be adapted at each timestep ˙ ϕ and ˙ α can be computed using the differential kinematic model (cf paper) The more the motion differs from a quasi-sinusoid, the more the power consumption increases

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Pseudo-periodical motion

To obtain a pseudo-periodical motion : ➔ modification of the quasi-sinusoidal motion The motor velocities have to be adapted at each timestep ˙ ϕ and ˙ α can be computed using the differential kinematic model (cf paper) The more the motion differs from a quasi-sinusoid, the more the power consumption increases

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Conclusion

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Conclusion

A multi-objective evolutionary algorithm has been used to determine, for a horizontal flight :

typical flight speed (10-12 m/s) angle ranges power required to fly (20-50 W/kg)

Simulations show that wing folding substantially decreases the required power (25-44%) These data have been used to design an efficient innovative parallel wing-beat mechanism

any kinematic is possible minimum energy consumption = sinusoidal movement

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Future work

This is only a preliminary work Basic aerodynamic measurements Whirling arm experiments :

design of control laws comparison of wing designs evolution of open-loop controllers evolution of neural-network controllers ...

Folding wings

Jean-Baptiste Mouret Robur project

Introduction Evolutionary optimization Mechanical design Conclusion Questions

Questions

Contact : jean-baptiste.mouret@lip6.fr This study was funded by a grant from Parinov

Jean-Baptiste Mouret Robur project