Team Description: Pumas@Home 2018
Affiliation: Bio-Robotics Laboratory, School of Engineering, National Autonomous University of Mexico (UNAM)
Participation: RoboCup@Home 2018 & ROCKIn — Open Platform League (OPL)
Team Members: Jesús Savage, Reynaldo Martell, Marco Negrete, Julio Cruz, Jesús Cruz, Hugo Estrada, José Cruz, Edgar Vázquez, Jaime Márquez, Edgar Silva, Manuel Pano, Luis Álvarez, and Mauricio Matamoros.
1. About the Team & Track Record
The UNAM Bio-Robotics Laboratory has competed continuously in RoboCup@Home since Bremen 2006. Notable achievements include 3rd place in Atlanta 2007, world finals in 2014 and 2015, and 4th place along with the Best in Speech Recognition and Natural Language Understanding award in Nagoya 2017.
2. Software Architecture: ViRBot System
Justina's software operates under the layered ViRBot (Virtual and Real Robot) architecture, bridging Linux ROS nodes and Windows modules via our proprietary Blackboard middleware:
- Inputs Layer: Gathers raw sensor signals from LiDAR rangefinders, sonars, infrared, Kinect RGB-D cameras, and microphone arrays. Signal processing converts sensory streams into symbolic beliefs about the environment.
- Knowledge Management Layer: Maintains 2D SLAM occupancy maps, Kalman filter localization, and a rule-based expert system encoded with NASA's CLIPS.
- Planning Layer: Validates perceptual hypotheses against the knowledge base, recognizes situations, and determines physical operation sequences.
- Execution Layer: Deploys hardwired finite state machines executing base behaviors (person tracking, speech interaction, navigation, and object manipulation).
3. Key Research & Novel Contributions
A. Natural Language Understanding via Conceptual Dependency (CD)
Semantic action representation using Schank's Conceptual Dependency (CD) theory integrated into a rule-based expert system:
- Deconstructs sentences into conceptual primitives such as ATRANS (transfer of abstract relationships, e.g., give/take) and PTRANS (transfer of physical location, e.g., go).
- Extracts the Actor, Act, Object, Direction, and State without demanding a rigid grammatical structure.
- CLIPS inference engine evaluates production rules against CD facts to automatically synthesize deterministic state-space action plans.
B. Crowd Facial Detection via Panoramic Image Stitching
A multi-image stitching pipeline designed to count and analyze crowded groups across wide visual angles without single-frame FOV limitations:
- Captures multiple sequential RGB frames across varied pan angles.
- Extracts invariant visual landmarks using SURF (Speeded Up Robust Features) and matches points across frames with FLANN.
- Computes homography matrices iteratively via RANSAC to compensate for perspective distortions, lens calibration differences, and color transitions.
- Runs Haar Cascade facial detectors on the resulting panoramic image to identify standing, sitting, or lying individuals.
C. Sound Source Localization (SSL)
Direction of Arrival (DOA) estimation over an angular range of [0, 2π) using an equilateral triangular microphone array:
- Acquires and digitizes audio phase information via a dedicated Texas Instruments TMS320F28377S MCU with dual 12-bit ADCs sampling up to 1.1 MSPS.
- Calculates the Cross-Power Spectral Density (CPSD) between microphone pairs: G12(ω) = X1*(ω) X2(ω) = P(ω) e-jωτ.
- Computes an integrated cross-spectrum where the global maximum identifies the precise angle of arrival of the human speaker.
4. Robot Justina: Technical Specifications
| Mobile Base | Omnidirectional base driven by differential pair configuration and omnidirectional wheels (new design) |
| Torso & Head | 1-DOF elevating torso (worm screw and gear mechanism); 2-DOF pan-and-tilt head powered by Dynamixel servomotors |
| Manipulators | Dual 7-DOF anthropomorphic arms actuated by 10 Dynamixel servomotors each |
| Sensors & Cameras | Hokuyo URG-04LX-UG01 LiDAR, Microsoft Kinect RGB-D sensor, Logitech Pro C920 Full HD camera |
| Audio Hardware | Røde NTG2 directional microphone, 4-microphone sound source localization array, dual speakers |
| Software Stack | ROS Kinetic & Blackboard middleware, ROS Gmapping, A* navigation, Haar Cascades, OpenPoses, Loquendo TTS, Microsoft Speech Recognition, NASA CLIPS |
| Compute Architecture | Onboard laptop running Ubuntu 16.04 LTS and Windows 7 VM; dedicated TMS320F28377S MCU for audio DSP |