Team Description: Pumas@Home 2020
Affiliation: Bio-Robotics Laboratory, School of Engineering, National Autonomous University of Mexico (UNAM)
Participation: RoboCup@Home 2020 — Open Platform League (OPL)
Team Members: Jesús Savage, Julio Cruz, Reynaldo Martell, Hugo Estrada, Marco Negrete, Diego Cordero, Manuel Pano, Julio Martínez, and Luis González.
1. About the Team & Track Record
The UNAM Bio-Robotics Laboratory has competed continuously in RoboCup@Home since Bremen 2006. Pumas Team achieved 2nd place in both RoboCup 2018 and 2019, 4th place and the Best in Speech Recognition and Natural Language Understanding award in Nagoya 2017, along with strong performances in the European ROCKIn challenge.
2. Software Architecture: ViRBot System
Justina's software operates under the ViRBot (Virtual and Real Robot) architecture, bridging Linux ROS nodes and Windows modules via our custom Blackboard middleware:
- Inputs Layer: Reads raw data from LiDAR, sonar, infrared, microphones, and Kinect RGB-D sensors, transforming them via digital signal processing and pattern recognition into symbolic environmental beliefs.
- Knowledge Management Layer: Maintains SLAM occupancy maps, Kalman Filter localization, and an expert system based on NASA's CLIPS to store domain rules.
- Planning Layer: Validates perceptual beliefs against stored knowledge, identifies situational context, and generates sequences of physical operations to satisfy goals.
- Execution Layer: Deploys finite state machines for modular behaviors (such as person recognition, navigation, and object manipulation).
3. Key Research & Novel Contributions
A. Natural Language Understanding via Conceptual Dependency (CD)
Action representation using Schank's Conceptual Dependency (CD) theory within an expert system:
- Decomposes spoken instructions into conceptual action primitives such as ATRANS (transfer of abstract relations, e.g., give/take) and PTRANS (transfer of physical location, e.g., go).
- Extracts the actor, act, object, and direction to instantiate semantic facts directly into CLIPS production rules.
- Allows the inference engine to construct deterministic state-space action plans without requiring strict, rigid grammatical syntax.
B. Multi-Robot HRI via Smart Home Integration (Amazon Alexa)
A distributed human-robot interaction scheme allowing users to dispatch tasks to multiple domestic robots from anywhere in the home:
- Smart devices (Amazon Echo) capture voice requests and route them to Alexa Voice Service (AVS) customized with dedicated skills.
- A Node.js web-service node parses requests, generates corresponding Conceptual Dependencies, and bridges commands to Justina and Takeshi through ROS Bridge.
- Enables multi-robot cooperative tasks, such as Takeshi welcoming guests at the entrance while Justina serves drinks in the living area.
C. Semantic Segmentation via Mask R-CNN
Integration of Mask R-CNN running on onboard GPU hardware for advanced object detection, extracting regions of interest (RoI) to perform bounding-box regression and precise pixel-level mask generation in real time.
4. Robot Justina: Technical Specifications
| Mobile Base | Omnidirectional base with differential pair configuration and omnidirectional wheels |
| Torso & Head | 1-DOF elevating torso (worm screw and gear mechanism); 2-DOF pan-and-tilt head with Dynamixel servomotors |
| Manipulators | Dual 7-DOF anthropomorphic arms equipped with 10 Dynamixel servomotors each |
| Sensing & Vision | Hokuyo URG-04LX-UG01 LiDAR, Microsoft Kinect RGB-D sensor, Logitech Pro C920 Full HD camera |
| Audio Hardware | Røde NTG2 shotgun microphone, 4-microphone sound source localization array, dual speakers |
| Software Stack | ROS Kinetic & Blackboard middleware, ROS Gmapping, A* navigation, YOLO, FaceNet, OpenPose, Loquendo TTS, NASA CLIPS |
| Embedded Compute | NVIDIA Jetson TX2 Developer Kit for GPU-accelerated vision, running Ubuntu 16.04 LTS and Windows 7 VM |