Active MultiModel Control for UCAVs: Software mechanisms and infrastructure

1
Software Enabled ControlActive Multi-Model
Control of Uninhabited Combat Air
VehiclesHoneywell Technology CenterReview
MeetingSan Francisco, CA, October 20,
1999AFRL/DARPA Contract No. F33615-98-C-1340
HTC Team Mukul Agrawal Dan Bugajski Darren Cofer
(co-p.i.) Datta Godbole Vipin Gopal Jeff
Rye Tariq Samad (p.i.) Don Shaner Fred Wagener
2
Outline

• Overview
• Wavelet-based route generation
• Interior point optimization for maneuver control
• Inner loop flight control design
• demo
• Software infrastructure
• Multi-vehicle integrated simulation environment
• demo

Tariq Samad
Darren Cofer
3
Active Multi-Model Control
Mission Control Today
Software Enabled Control Vision
Long time horizon No dynamics Mission-level goals
Strategic
Strategic
Active Multi-Model Control as the enabling
technology
Intermediate horizon Approximate
dynamics Maneuver optimization
Tactical
Tactical
Immediate time horizon High-fidelity
dynamics Tracking, disturbance rejection
Supervisory
Supervisory
Strategic, tactical, supervisory control systems
developed and operated independently Models used
at all levels, but little attempt at cohesion or
consistency Suboptimal performance and agility,
and lack of autonomous capabilities

• Seamless mode transitions
• Agility and flexibility
• Uncertainty management for complex systems
• High-performance extreme maneuvers
• Coordinated formation flight
• Autonomy and intelligence

4
Multi-resolution Flight Control
Resolution for models of terrain/threat/weather
Vehicle models
Technology
Space-Time preview
Update rate
Evolutionary computing (Wavelet basis)
Maneuver Generation
Curves in 3D Turn constraints
coarse
Full
minutes
yd
3 DOF point-mass state and i/p constraints curve-f
it for aero-data
Nonlinear Optimization (Wavelet basis)
Maneuver Optimization
Next few way-points
variable
seconds
xd,ud
(Decoupled) 6DOF model state and i/p constraints
Maneuver regulation based on Nonlinear Hybrid
System Theory
fine
Trajectory segment
40 Hz
Maneuver Regulation
Actuator commands
None
continuous
Aircraft
5
Challenge Problem
Dynamic re-optimization of UCAV missions as
unforeseen situations arise - focus on real-time
and high- performance control
6
Requirements for Autonomous Control

• What does autonomy imply?
• appropriate reactions to unforeseen situations
• adaptation of planned activities to current
  environment
• coordination with other agents (vehicles, humans)
• Some implications
• all control behaviors cannot be pre-compiled
• an autonomous system must have knowledge--of
  itself, its surroundings, its objectives, its
  friends and enemies, etc.
• Key requirement Active Multi-Models
• multi-models diverse knowledge sources are
  necessary
• active models will need to be executed on-line

7
Wavelet Representation
Scaling function
Wavelet
Quadratic spline wavelet - compact support -
differentiable - closed-form solution
Scale, m
am,n, bm,n, cn, dn define a rich space
for (x(t),y(t)) trajectories
Translation, n
8
Multi-Scale Trajectory Optimziation
(simplified representation)

• Define x, y at higher resolution for immediate t,
  at successively coarser resolution for
  increasingly future t
• Expend computational resources in proportion to
  resolution needed

0
0
0
0
0
0
0
Scale, m frequency
0
0
0

• Set am,n, bm,n to 0 selectively

• Optimize non-zero parameters

0

• Multi-resolution models for aircraft, terrain
  needed for computing cost, checking constraints

Translation, n time

• Can dynamically bring in or remove resolution
  levels at various future points, as projected
  needs dictate

9
Wavelet-Based Trajectory Optimization Example

• Baseline route straight line between Current
  and End Point
• At Current point, new information about threat,
  target received
• Route optimized dynamically with evolutionary
  computing algorithm
• Nonlinear model predictive control
  framework--repeated incremental optimization as
  needed
• Orthogonal/semi-orthogonal wavelet bases
  facilitate rapid refinement, adjustment

10
Interior Point Methods for Maneuver Optimization
Waypoints
3 degrees of freedom f16 model used f(x)
g(x,u)
Solution with IrSQP
Trajectory is

• Consistent with f16 dynamics
• Flyable
• Optimal with respect to a chosen cost/time
  criterion

Generation of optimal trajectory
11
Aircraft model for Maneuver Optimization

• 3 DOF point-mass model with constraints
• states , inputs
• Maneuver optimization produces
• reference state and input trajectories for
  regulation

12
Trajectory Optimization--irSQP
Interior point reduced Hessian Successive
Quadratic Programming (IrSQP)

• Solution nuggets
• Interior point methods to better handle difficult
  inequality constraints
• Computations in the reduced space of degrees of
  freedom to facilitate faster solutions
• Decomposition mechanisms to exploit the problem
  structure

• Features
• Solution satisfies a rigorous mathematical
  optimality criterion.
• Easy incorporation and natural treatment of
  constraints
• Example - forbidden areas of flight
• Flexible optimization capabilities - Optimize
  with respect to the criterion chosen at a given
  time
• Example - min time, min fuel

13
IrSQP and the F-16 Model
F-16 formulation
NLP
Minimum fuel/control/time
Discretized differential equations for 3 dof
model (x,y,h,V,g,c,P)
Bounds on state and control variables, Restricted
regions of flight
Solution by a sequence of quadratic program
approximations at the current iterate
QP
Interior point methods solve the optimality
conditions directly
Optimality Conditions
14
Interior Point Iterates
Predictor Step
rs are constraint residuals
Corrector Step
Second order correction
Centering term
15
Conflict Resolution Application
To compute optimal trajectories that resolve
potential conflicts among multiple aircraft
Conflict avoidance constraints more difficult to
handle as the number of aircraft increases
Solution with novel interior point
methods. Rigorous mathematicalapproach -
viable for any number of aircraft
OriginalTrajectories
16
Maneuver Regulation

• Maneuver Regulation
• follow optimal trajectory in the presence of
  disturbances
• maintain safety
• Approach
• define a parameterized set of basic maneuvers
• design individual feedback controllers for
  regulation of each maneuver
• design switching logic to
• maintain safety
• minimize tracking errors
• handle fault management
• Can easily incorporate multi-UAV coordination

6DOF Aircraft Model
Maneuver Regulation (Dynamic Inversion)
xd ud yd
y
Actuator commands
error
Trajectory Optimization
17
Upcoming Research Tasks

• Extend the multi-resolution algorithm to
• multi-UAV analysis
• parallel executions across multiple UAVs
• Coordinated control for multi-aircraft missions
• design of communication language and protocols
• coordinated control algorithms for different
  maneuvers
• formation flying
• group evasive maneuvers
• safety watchdog
• Is the aircraft operating within its safety
  envelope?
• Fault management
• verification and performance evaluation of
  control algorithms

18
New Coordinated Control Concepts
Modes as parametrized trajectories
Hybrid safety watchdog controller
Maneuver controller
19
Innovations to date

• Multi-resolution representations for UAV maneuver
  generation
• Evolutionary computing route optimization
  algorithm
• Interior point optimization for extreme
  performance control
• Adaptive, dynamic resource allocation for UAV
  fleet
• Multi-vehicle software simulation/evaluation
  environment
• Seamless mode transition mechanisms integrating
  above elements

Exciting prospects for new control technology and
autonomous systems
20
Active Multi-Model Control for UCAVs Software
mechanisms and infrastructure

• Q What software infrastructure is needed to
  enable the implementation of reliable and
  predictable control applications based upon
  multiple interacting dynamic models?
• (a.k.a. Active Multi-Model Control)

21
Premise

• Integrated architectures (vs. federated) support
  reactive modification of execution structure
• allocation of CPU time
• hardware/software binding

• fault tolerance
• etc.

f1
f1
f1
more f2
Application requirements
f2
f2
f2
less f1
f3
f3
Integrated generic hardware
Federated dedicated hardware
22
Requirements for SW infrastructure

• Goal
• Maximize performance of control software given
  constraints of HW platform capabilities and
  required response times.
• Capabilities
• CPU resource allocation
• application-triggered adaptation
• task specification of service needs
• schedulability model
• resource manager to perform adaptations
• coherent sharing of state information

23
Control view
select model fidelity / algorithm
generate maneuver(s)
turn-angle model
kinematic A/C model
dynamic A/C model
possible maneuvers
flyability constraint
change update rate
constraints ? fitness ?
maneuver control
threat model
models constraints
threat constraint
cost(s)
change resolution
select/enact maneuver
terrain constraint
terrain model
targets control modes
autopilot
sensors/nav
vehicle control
state
commands
UCAV
24
Computational view
autopilot
maneuver gen
maneuver eval
maneuver select
maneuver select
flyability constraint
threat constraint
terrain constraint
dynamic A/C model
threat model
terrain model
tasks
? service requests service guarantees, execution
schedule ?

• Service parameters
• CPU load (exec time)
• rate
• deadline
• criticality
• etc.

schedulability model
object distribution
resource manager
software infrastructure
I/O
comm
RTOS
sensors
actuators
networks
CPU
Hardware
25
Model characteristics

• more iterations
• more maneuvers
• longer horizon
• higher resolution
• higher fidelity model
• more samples

What would a task do with more time?
26
Adaptation

• How is adaptation controlled?
• Based on computed / observed state, set task
  criticality and computing requirements.
• CPU resource (rate x load) is made available to
  tasks based on criticality, requests, and
  schedulability analysis.
• Control tasks execute with allotted time. Adapt
  to meet application constraints (deadlines,
  accuracy).

Heres what I want.
Heres what you get.
This is how Ill use it.
27
Example mechanisms
generate maneuver(s)
candidate maneuvers for evaluation
Buffer
Solution space

• Computing resource allocation
• Equal ? complete evaluation of each new maneuver
• Weighted ? quickly rule out infeasible maneuvers
  using most stringent criterion
• Data sharing (eval. state)
• Evaluator selects best maneuver so far

evaluate flyability
evaluate terrain
evaluate weather
select/enact maneuver
28
Simulation environment

• Goal
• Integrate/test control algorithms SW
  infrastructure
• Requirements
• control real executable code
• detailed A/C models
• multiple A/C, separate processes
• real-time performance data
• NT platform

Active multi-model control (real-time)
Aircraft dynamics (simulated)
World state object interactions and communication
29
Simulation architecture
wait for ltworld state readygt
initialize
A/C in separate processes
filter world state (inputs)
local observation, communication, time
ltworld state readygt
real-time control
20 mSec real time (scaled for performance)
A/C 1
A/C 2
A/C n
ltA/C 1 rdygt
ltA/C 2 rdygt
ltA/C n rdygt
reconcile world state
simulate aircraft dynamics (outputs)
20 mSec simulated time (50 Hz A/C model)
control algorithms in separate threads
save
display
signal ltA/C readygt
30
Operational scenario

• UAV mission
• cruising to target, wind optimal
• updated threat trajectory at t
• generate maneuver to avoid
• enter terrain following mode
• Adaptation required
• evasive maneuver (response by t ?)
• increase threat tracking rate
• higher fidelity flyability model
• decrease resources devoted to other models
• (weather updates, terrain resolution,)

31
Demo

• RegControl low-level control loops
• constant time, highest criticality
• RouteGen produce candidate maneuvers for
  evaluation
• adapt rate
• FlyModel test maneuver for flyability
• adapt algorithm and resolution
• WxModel prediction and input for wind/weather
  optimality
• adapt update rate, lowest criticality
• ThreatModel track threats
• adapt update rate
• TerrainModel provide bubble of detailed
  interpolated map data
• adapt size and resolution

32
Summary

• Use UAV scenarios models to derive SW
  infrastructure requirements
• Ongoing Prototype resource allocation
  adaptation mechanisms for AMMC
• Capitalize on related efforts
• RTARM
• MetaH scheduler / slack scheduling
• Crusader operating environment