experimental comparison of uncertainty criteria for active slam

21
Experimental Comparison of Uncertainty Criteria for Active SLAM Henry Carrillo

Upload: holland

Post on 21-Jan-2016

44 views

Category:

Documents


0 download

DESCRIPTION

Experimental Comparison of Uncertainty Criteria for Active SLAM. Henry Carrillo. Motivación. Preliminares. Experimentos. Primer experimento : acerca del calculo Segundo experimento : SLAM activo. Robot simulado ambiente interior : MRPT / C++ - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Experimental Comparison of Uncertainty Criteria for Active SLAM

Experimental Comparison of Uncertainty Criteria for

Active SLAM

Henry Carrillo

Page 2: Experimental Comparison of Uncertainty Criteria for Active SLAM

Motivación

2

Page 3: Experimental Comparison of Uncertainty Criteria for Active SLAM

Preliminares

3

Page 4: Experimental Comparison of Uncertainty Criteria for Active SLAM

Experimentos Primer experimento : acerca del calculo

Segundo experimento : SLAM activo

4

• Robot simulado ambiente interior : MRPT / C++

• Robot real ambiente interior : Pioneer 3 DX - ad-hoc

• Robot real ambiente interior : DLR dataset• Robot real ambiente exterior : Victoria Park

dataset• Robot simulado con horizonte unitario : MRPT / C++

Page 5: Experimental Comparison of Uncertainty Criteria for Active SLAM

Primer experimento Primer experimento : acerca del calculo

Es posible calcular D-opt en un robot realizando SLAM?

Ejecutamos un algoritmo de SLAM "secuencial " (e.g EKF-SLAM, iSAM)

Calculamos en cada paso : A-opt, E-opt , D-opt, det. de la covarianza, entropía e información mutua.• Robot simulado ambiente interior : MRPT /

C++• Robot real ambiente interior : Pioneer 3 DX

- ad-hoc• Robot real ambiente interior : DLR dataset• Robot real ambiente exterior : Victoria Park

dataset5

Page 6: Experimental Comparison of Uncertainty Criteria for Active SLAM

1E - Robot simulado ambiente interior (I)

Escenario: Área de 25x25m 2D EKF-SLAM Sensor: Odometria

+cámara (360 - 3m rango)

180 landmarks - DA conocida

Errores Gaussianos: Odometria + sensores

6

Page 7: Experimental Comparison of Uncertainty Criteria for Active SLAM

1E - Robot simulado ambiente interior (II)

(a)-(f) A-opt, E-opt, D-opt, determinant, entropy and MI7

Page 8: Experimental Comparison of Uncertainty Criteria for Active SLAM

1E-Robot en ambiente interior @ DLR (I)

Escenario: Oficina 60x40 m 2D EKF-SLAM Sensor: Odometria + cámara

BW 576 landmarks – DA

conocida

8

Page 9: Experimental Comparison of Uncertainty Criteria for Active SLAM

1E-Robot en ambiente interior @ DLR (II)

(a)-(f) A-opt, E-opt, D-opt, determinant, entropy and MI9

Page 10: Experimental Comparison of Uncertainty Criteria for Active SLAM

1E-Robot en ambiente exterior @ VP (I)

Escenario: Parque de 350 x 350 m iSAM Sensor: Odometria +

Laser xxx landmarks – DA

conocida

10

Page 11: Experimental Comparison of Uncertainty Criteria for Active SLAM

1E-Robot en ambiente exterior @ VP (II)

(a)-(f) A-opt, E-opt, D-opt, determinant, entropy and MI11

Page 12: Experimental Comparison of Uncertainty Criteria for Active SLAM

1E-Robot en ambiente interior ad-hoc (I)

Escenario: Área de 25x25m 2D EKF-SLAM Sensor: Odometria

+cámara (360 - 3m rango) 180 landmarks - DA

conocida Errores Gaussianos: Odometria + sensores

12

Page 13: Experimental Comparison of Uncertainty Criteria for Active SLAM

1E-Robot en ambiente interior ad-hoc (II)

(a)-(f) A-opt, E-opt, D-opt, determinant, entropy and MI13

Page 14: Experimental Comparison of Uncertainty Criteria for Active SLAM

Segundo experimento Segundo experimento : SLAM activo

Que efecto tiene la métrica de incertidumbre en el SLAM activo?

SLAM activo == Horizonte unitario (greedy), discreto.

Métricas de incertidumbre == A-opt, D-opt y entropía.

Efecto == MSE y • Robot simulado con horizonte unitario : MRPT / C++

14

Page 15: Experimental Comparison of Uncertainty Criteria for Active SLAM

2E-Robot en ambiente interior ad-hoc (I)

Escenario: Área de 20x20m y

30x30m 2D EKF-SLAM Sensor:

Odometria+cámara (360 - 3m rango)

15

Errores Gaussianos: Odometria+sensores

Planeador de caminos: Discreto (A*) y continuo (Atraccion-Repulsion)

Page 16: Experimental Comparison of Uncertainty Criteria for Active SLAM

2E-Robot en ambiente interior ad-hoc (II)

Resulting paths from each uncertainty metric: (a) D-opt, (b) A-opt and (c) Entropy. Each colour represents an executed path. The planning area was 20 x 20 m.

16

Page 17: Experimental Comparison of Uncertainty Criteria for Active SLAM

2E-Robot en ambiente interior ad-hoc (III)

Resulting trajectories for a 10000 steps active SLAM simulation. (a). Predefined trajectory and landmarks ground truth. (b). A-opt based active SLAM. (c). D-opt based active SLAM.

17

Page 18: Experimental Comparison of Uncertainty Criteria for Active SLAM

2E - Análisis cuantitativo 30x30 m

Evolution of the MSE ((a)-(c)) and 2 ((d)-(f)). Average of 10 MC.

18

Page 19: Experimental Comparison of Uncertainty Criteria for Active SLAM

2E - Análisis cuantitativo 20x20 m

Evolution of the MSE ((a)-(c)) and 2 ((d)-(f)). Average of 10 MC.

19

Page 20: Experimental Comparison of Uncertainty Criteria for Active SLAM

Take home message

20

Page 21: Experimental Comparison of Uncertainty Criteria for Active SLAM

Experimental Comparison of UncertaintyCriteria for Active SLAM

21

Gracias!!!