Introduction to RoboCup@Home
1 like462 views
The document discusses RoboCup@Home, a competition for domestic service robots that focuses on human-robot interaction and mobile manipulation. It describes some of the standardized tasks robots must complete, including delivering drinks in a cocktail party setting and retrieving objects in an unknown environment like a restaurant. The document also discusses challenges like navigation, manipulation of everyday objects, and interacting with humans in noisy environments. Finally, it provides examples of how imitation learning and learning unknown words could help robots perform household activities.
![LCore Applied (2): Learning unknown words
• Difficulty: low phoneme recognition accuracy
• Proposed
– Learns phoneme sequence with waveform
– Voice conversion using EigenVoice Gaussian Mixture Model*[Toda+ 2007]
Evaluation using CMOS metric
• proposed method outperformed baseline
9](https://image.slidesharecdn.com/20131213robocupathome-131213044502-phpapp02/85/Introduction-to-RoboCup-Home-9-320.jpg)
Introduction to RoboCup@Home
- 1. Introduction to RoboCup@Home Imitation learning applied to domestic service robot tasks 2013/12/13 Komei Sugiura National Institute of Information and Communication Technology, Japan komei.sugiura@nict.go.jp
- 2. RoboCup@Home: Benchmark tests for domestic robots • RoboCup@Home: The largest competition for domestic robots – One of the major RoboCup leagues – Focuses on human-robot interaction and mobile manipulation – Robots are evaluated by 7 standardized and 3 demonstration tasks • Info – >200 participants from 15 countries – 6-10 members/team 2
- 3. 3
- 4. Difficulties • Mobile manipulation – Navigation in unknown environments – Surrounded by spectators – Real shop environments – Manipulation of everyday objects • Human robot interaction – Very noisy environments – Robust dialogue management – Gesture recognition 4
- 5. Standard test 1: Cocktail Party • Task: learn and recognize unknown persons, and deliver drinks Item Max score Best team Average Detecting the calling persons 150 x 3 300 95 Understanding human/drink names 100 x 3 300 108 Delivering correct ordered drinks 200 x 3 400 31.6 5
- 6. Standard test 2: Restaurant • Task: Retrieve three objects in an unknown environment e.g. restaurant • Environment: a real restaurant (robots are transported) Item Score Best team Average Reaching a location in the guide phase 50 x 5 250 143 Reaching a location in the navigation phase 100 x 4 200 50 Grasping the correct objects 250 x 3 500 45 6
- 7. Standard test 3: Enduring General Purpose Service Robots • NimbRo (Bonn University)
- 8. LCore Applied (1): Imitation learning for household activities Teacher: “Throw-into.” ( with demonstrating motions several times) Robot: (Estimates relative objects and learns motion trajectories) User: “Throw a plastic bottle into a dust bin.” Robot: (Searches for the objects and executes “throw-into” motion) Dialog example
- 9. LCore Applied (2): Learning unknown words • Difficulty: low phoneme recognition accuracy • Proposed – Learns phoneme sequence with waveform – Voice conversion using EigenVoice Gaussian Mixture Model*[Toda+ 2007] Evaluation using CMOS metric • proposed method outperformed baseline 9