roboticsaima.eecs.berkeley.edu/slides-pdf/chapter25.pdf · robotics chapter 25 chapter 25 1....
TRANSCRIPT
Robotics
Chapter 25
Chapter 25 1
Outline
Robots, Effectors, and Sensors
Localization and Mapping
Motion Planning
Motor Control
Chapter 25 2
Mobile Robots
Chapter 25 3
Manipulators
R
RRP
R R
Configuration of robot specified by 6 numbers⇒ 6 degrees of freedom (DOF)
6 is the minimum number required to position end-effector arbitrarily.For dynamical systems, add velocity for each DOF.
Chapter 25 4
Non-holonomic robots
θ
(x, y)
A car has more DOF (3) than controls (2), so is non-holonomic;cannot generally transition between two infinitesimally close configurations
Chapter 25 5
Sensors
Range finders: sonar (land, underwater), laser range finder, radar (aircraft),tactile sensors, GPS
Imaging sensors: cameras (visual, infrared)Proprioceptive sensors: shaft decoders (joints, wheels), inertial sensors,force sensors, torque sensors
Chapter 25 6
Localization—Where Am I?
Compute current location and orientation (pose) given observations:
Xt+1Xt
At−2 At−1 At
Zt−1
Xt−1
Zt Zt+1
Chapter 25 7
Localization contd.
xi, yi
vt ∆t
t ∆t
t+1
xt+1
h(xt)
xt
θt
θ
ω
Z1 Z2 Z3 Z4
Assume Gaussian noise in motion prediction, sensor range measurements
Chapter 25 8
Localization contd.
Can use particle filtering to produce approximate position estimate
Robot positionRobot position
Robot position
Chapter 25 9
Localization contd.
Can also use extended Kalman filter for simple cases:
robot
landmark
Assumes that landmarks are identifiable—otherwise, posterior is multimodal
Chapter 25 10
Mapping
Localization: given map and observed landmarks, update pose distribution
Mapping: given pose and observed landmarks, update map distribution
SLAM: given observed landmarks, update pose and map distribution
Probabilistic formulation of SLAM:add landmark locations L1, . . . , Lk to the state vector,proceed as for localization
Chapter 25 11
Mapping contd.
Chapter 25 12
3D Mapping example
Chapter 25 13
Motion Planning
Idea: plan in configuration space defined by the robot’s DOFs
conf-3
conf-1conf-2
conf-3
conf-2
conf-1
w
w
elb
shou
Solution is a point trajectory in free C-space
Chapter 25 14
Configuration space planning
Basic problem: ∞d states! Convert to finite state space.
Cell decomposition:divide up space into simple cells,each of which can be traversed “easily” (e.g., convex)
Skeletonization:identify finite number of easily connected points/linesthat form a graph such that any two points are connectedby a path on the graph
Chapter 25 15
Cell decomposition example
start
goal
startgoal
Problem: may be no path in pure freespace cellsSolution: recursive decomposition of mixed (free+obstacle) cells
Chapter 25 16
Skeletonization: Voronoi diagram
Voronoi diagram: locus of points equidistant from obstacles
Problem: doesn’t scale well to higher dimensions
Chapter 25 17
Skeletonization: Probabilistic Roadmap
A probabilistic roadmap is generated by generating random points in C-spaceand keeping those in freespace; create graph by joining pairs by straight lines
Problem: need to generate enough points to ensure that every start/goalpair is connected through the graph
Chapter 25 18
Motor control
Can view the motor control problem as a search problemin the dynamic rather than kinematic state space:
– state space defined by x1, x2, . . . , x1, x2, . . .
– continuous, high-dimensional (Sarcos humanoid: 162 dimensions)
Deterministic control: many problems are exactly solvableesp. if linear, low-dimensional, exactly known, observable
Simple regulatory control laws are effective for specified motions
Stochastic optimal control: very few problems exactly solvable⇒ approximate/adaptive methods
Chapter 25 19
Biological motor control
Motor control systems are characterized by massive redundancy
Infinitely many trajectories achieve any given task
E.g., 3-link arm moving in plane throwing at a targetsimple 12-parameter controller, one degree of freedom at target11-dimensional continuous space of optimal controllers
Idea: if the arm is noisy, only “one” optimal policy minimizes error at target
I.e., noise-tolerance might explain actual motor behaviour
Harris & Wolpert (Nature, 1998): signal-dependent noiseexplains eye saccade velocity profile perfectly
Chapter 25 20
Setup
Suppose a controller has “intended” control parameters θ0
which are corrupted by noise, giving θ drawn from Pθ0
Output (e.g., distance from target) y = F (θ);y
Chapter 25 21
Simple learning algorithm: Stochastic gradient
Minimize Eθ[y2] by gradient descent:
∇θ0Eθ[y
2] = ∇θ0
∫Pθ0
(θ)F (θ)2dθ
=∫ ∇θ0
Pθ0(θ)
Pθ0(θ)
F (θ)2Pθ0(θ)dθ
= Eθ[∇θ0
Pθ0(θ)
Pθ0(θ)
y2]
Given samples (θj, yj), j = 1, . . . , N , we have
ˆ∇θ0Eθ[y2] =
1
N
N∑j=1
∇θ0Pθ0
(θj)
Pθ0(θj)
y2
j
For Gaussian noise with covariance Σ, i.e., Pθ0(θ) = N(θ0, Σ), we obtain
ˆ∇θ0Eθ[y2] =
1
N
N∑j=1
Σ−1(θj − θ0)y2
j
Chapter 25 22
What the algorithm is doing
x
xx
x x
xx
x xx
xx
xx x x
xxxx
xx
x
xxxx
xxx
x
xx
x
xx
x
x xx
xxx
Chapter 25 23
Results for 2–D controller
4
5
6
7
8
9
10
0.2 0.4 0.6 0.8 1 1.2
Vel
ocity
v
Angle phi
Chapter 25 24
Results for 2–D controller
4.51
4.52
4.53
4.54
4.55
4.56
4.57
4.58
4.59
4.6
4.61
0.6 0.61 0.62 0.63 0.64 0.65
Vel
ocity
v
Angle phi
Chapter 25 25
Results for 2–D controller
0.0055
0.006
0.0065
0.007
0.0075
0.008
0.0085
0.009
0.0095
0 2000 4000 6000 8000 10000
E(y
^2)
Step
Chapter 25 26
Summary
The rubber hits the road
Mobile robots and manipulators
Degrees of freedom to define robot configuration
Localization and mapping as probabilistic inference problems(require good sensor and motion models)
Motion planning in configuration spacerequires some method for finitization
Chapter 25 27