7. SpCoSLAM [Taniguchi 17]
マルチモーダル情報に基づくノンパラメトリックベイズモデルとSimultaneous Localization
And Mapping (SLAM) 、語彙獲得の統合モデル
[Taniguchi 17] Taniguchi, A., et al. : Online Spatial Concept and Lexical Acquisition with Simultaneous Localization and Mapping, IEEE/RSJ IROS, pp. 811-818 (2017)
地図
位置分布
(ガウス分布)
自己位置推定
1
t
x
xt
xt 1
t
z
t
u
t
C
t
i
l
0
0,
m
0
0,
V
1
t
z
1
t
u
1
t
z
1
t
u
LM
t
y
AM
t
S
l
t
f
∞
∞
m
l
W
k
k
SLAM
場所
画像
単語
音声認識・語彙獲得
マルチモーダル
カテゴリゼーション “Third table”
“Meeting space”
7
8. 場所概念のオンライン学習
• 未知環境下からの地図・場所概念・語彙の逐次学習が可能
– Rao-Blackwellized Particle Filter に基づくオンライン学習アルゴリズムをSpCoSLAMの
確率的生成モデルに適用
[Taniguchi 20] Taniguchi, A., et al. : "Improved and Scalable Online Learning of Spatial Concepts and Language Models with Mapping",
Autonomous Robots, 44(6), 927-946, 2020.
8
23. Spatial Concept-based Topometric Semantic Mapping for
Hierarchical Path-planning from Speech Instructions
(Preprint)
https://arxiv.org/abs/2203.10820
23
Hierarchical spatial representation
provides a mutually understandable
form for humans and robots to
render language-based navigation
tasks feasible.
24. 1
t
x
xt
xt 1
e
C
e
i
LM
e
y AM
e
S
l
e
f
𝐿
l
W
t
z
t
u
1
t
z
1
t
u 1
t
z
1
t
u
m
t
o
1
t
o 1
t
o
0
0,
m
0
0,
V
k
k
T
o
E
C
E
i
E
y
E
S
E
f
T
z
T
u
T
x
l
k
𝐾
Hidden semi-Markov model
(Topological level)
𝑖𝑒: Index of place
𝑓𝑒: Image feature
𝑆𝑒: Words
Speech recognition
Word segmentation
Multimodal mixture model
(Semantic level)
𝑂𝑡: Event-driven
variable
𝑢𝑡: Action variable
𝑦𝑒: Speech signal
𝑧𝑡: Depth sensor data
Simultaneous localization
and mapping (SLAM)
(Metric level)
SpCoTMHP: spatial concept-based topometric semantic mapping for
hierarchical path planning
24
27. Experiment II: 実験結果
Normalized mutual information (NMI) and adjusted Rand index (ARI) are the most widely used in clustering
tasks for unsupervised learning. 27
29. CONCLUSIONS
• We realized topometric semantic mapping and hierarchical
path planning
– with place-transition patterns based on multimodal observations by
robots.
• The navigation experiment using human speech instruction
– showed that the proposed path planning improves the performance
and reduces the calculation cost compared with conventional
methods.
29
32. zzzzzzzzzzzzzzz
We aim to improve the efficiency of place concept formation by autonomous active search
of mobile robots.
• Reducing the workload of users
• Improving learning accuracy by searching to reduce uncertainty
Active exploration of robots
• Decide where to ask for the name of the place and move to that position
Purpose
“tv” “sofa”
“entrance”
“kitchen”
32
33. zzzzzzzzzzzzzzz
場所概念形成の確率的生成モデルにおいて,パーティクルフィルタよる
オンライン学習と情報利得(Information Gain; IG) 最大化に基づく能動
的な探索を組み合わせた能動学習手法 SpCoAE を提案
SpCoAE:
移動先決定
移動
マルチモーダル
情報の収集
場所概念の
オンライン学習
ロボットによる
能動的な探索
ロボットの位置
+
ユーザによる
名前の発話
Spatial Concept Formation with
Information Gain-based Active Exploration
IGが最大となる探索候補点を選択
オンライン学習による場所概念の形成
観測する位置の選択
“This place is
living room.”
Word
information
Position
information
𝑥, 𝑦 = (1.8, −2.0)
能動的な探索による場所概念形成の流れ
この場所の名前を
教えてください
33
41. これまでの研究:動画の関連文献
• SpCoSLAM:場所概念のオンライン学習
– Akira Taniguchi, Yoshinobu Hagiwara, Tadahiro Taniguchi, and Tetsunari Inamura, "Online Spatial Concept and Lexical Acquisition
with Simultaneous Localization and Mapping", IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS2017),
pp.811-818, Sep, 2017.
– Akira Taniguchi, Yoshinobu Hagiwara, Tadahiro Taniguchi, and Tetsunari Inamura, "Improved and Scalable Online Learning of
Spatial Concepts and Language Models with Mapping", Autonomous Robots, 44(6), 927-946, Feb. 2020.
• 場所概念に基づく曖昧な命令からの推論
– Shota Isobe, Akira Taniguchi, Yoshinobu Hagiwara, and Tadahiro Taniguchi, "Learning Relationships between Objects and Places
by Multimodal Spatial Concept with Bag of Objects", International Conference on Social Robotics (ICSR), pp.115-125, Nov, 2017.
• ロボットの内部知識の可視化
– L. El Hafi, S. Isobe, Y. Tabuchi, Y. Katsumata, H. Nakamura, T. Fukui, T. Matsuo, G. A. Garcia Ricardez, M. Yamamoto, A. Taniguchi,
Y. Hagiwara, and T. Taniguchi, "System for Augmented Human-Robot Interaction Through Mixed Reality and Robot Training by
Non-Experts in Customer Service Environments", Advanced Robotics, Vol. 34, No. 3, Feb. 2020.
• 場所概念に基づくセマンティックマッピング
– Yuki Katsumata, Akira Taniguchi, Yoshinobu Hagiwara, and Tadahiro Taniguchi, "Semantic Mapping Based on Spatial Concepts for
Grounding Words Related to Places in Daily Environments", Frontiers in Robotics and AI, Vol. 6, No. 31, 2019.
– Y. Katsumata, A. Taniguchi, L. El Hafi, Y. Hagiwara, and T. Taniguchi, "SpCoMapGAN: Spatial Concept Formation-based Semantic
Mapping with Generative Adversarial Networks", IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2020),
pp. 1-8, Online, Oct. 2020.
• 場所概念の学習と他の環境への転移
– Yoshinobu Hagiwara, Keishiro Taguchi, Satoshi Ishibushi, Akira Taniguchi, Tadahiro Taniguchi, "Hierarchical Bayesian Model for the
Transfer of Knowledge on Spatial Concepts based on Multimodal Information", (Accepted to Advanced Robotics), 2021.
動画コンテンツ
• https://youtube.com/playlist?list=PLau2XPzh0eZnjO9a22RehimVt5gn1j_Lx
• https://youtube.com/playlist?list=PLDyf5Cyt8JrMgJhd8JetvNDSNvnBDOI_J
• https://emlab.jimdofree.com/multimedia/ 41