We are introducing Ellipso, a new domestic robot from FRACH robotics and we would like to hear your feedback!
Ellipso is an autonomous intelligent home robot, which is always online, listening to you and ready to execute your requests or just connect you with your loved ones. The robot implements an animal like brain model, so it behaves almost like it is alive. Ellipso is also integrated with Google Assistant, Google Drive, Google Mail, YouTube and Microsoft Skype to let you stay in touch.
Technical design details:
- Robot SW design - The robot's main control process is implemented as an Android long running background Service. The Service's thread cycles every 100ms and it implements integration with all external modules; such as camera, BlueTooth or Android's speech recognition API. The external modules translate their specific custom events to a generic logical language so the robotic brain model can simply analyse and command the robot based on the historical context and current events.
- Robot brain model - our robot brain cortex behavioral rules model is based on the “Affordance Competition Hypothesis” research of University of Montreal. This enables the robot to behave like it is alive and us to simply add new rules and behavior with minimal impact on the current ones. The same way a child would learn new behavioral patterns.
- Robot Android AI frameworks - Google TensorFlow Light deep neural networks for object detection (20 class yolo); Google FireBase ML framework for face/expression detection; OpenCV open sourced framework for face recognition.
- Robot HW AI frameworks - HW accelerated Kendryte K210 deep neural network object detection and sound FFT to detect specific sound patterns (e.g. human voice or a fire alarm).
- Robot SLAM mapping and navigation - "Fast Semi-Direct Monocular Visual Odometry" algorithm with IMU support to detect 3D obstacles and falls based on research of University of Zurich; simultaneous localisation and mapping (SLAM) algorithm for robot's self-navigation and map generation.
- Robot HW capabilities - Android ARM quad/octa core 1300/2000 MHz CPU for high level cognitive behavior and integration with Cloud services; Kendryte K210 dual core 400/600 MHz CPU with dedicated deep neural networks HW acceleration (TensorFlow Lite) and FFT acceleration; ESP32 dual core 240 MHz CPU for Bluetooth, UART and GPIO connectivity.
- Robot Android APIs - robot SW functionality and its behavior is tightly integrated with Android available public APIs and services. Such as Google Android Speech Recognition API; Google Android TextToSpeach API; Google Assistant and Google Youtube integration; Google Drive as Cloud based document storage; Google Mail as Cloud based communication API; Microsoft Skype as Cloud based IM service.
- Robot sensors - 32/16/5MPx upward and 2MPx downward facing two cameras; GPS satellite position detection, two microphones with HW accelerated FFT; two 3W speakers; two Doppler radar modules for movement detection; red 5mW laser pointer, 3W RGB LED lights, 3W white LED lights, 6 DoF IMU with gyros and accelerators.
- External connectivity - WiFi, Bluetooth, Android Cloud APIs. Our robots require only a standard Internet connectivity to access Google Mail, Google Drive and Microsoft Skype. No requirement to reconfigure your home or office WiFi routers to expose the IP ports or any other hassle.
- SW API dependency - our robots are only dependent on publicly available free Google and Microsoft APIs. Therefore, there is no risk to the robots functionality if anything happens to our company servers.
- Online SW updates - the robot's Android application (including its main behavioral rules and external interfaces) will be online updated every 3 months for new functionality and bug fixes; using Google Play market.
- No monthly charges - there are no monthly or other charges for using our robots. Have a fun and enjoy the robot every day.
- Self-charging - our robots automatically monitor status of their batteries and when the charge is low the robot will automatically locate and recharge itself in a docking station. The robot can operate about 4h on a single charge when actively working; about 12h on standby with radar movement detection enabled.
- Obstacle and voids detection - when on a table or facing down a staircase the robot will automatically detect dangerous voids (using the visual odometry algorithm) and reverse back.
- User replaceable batteries - users can simply replace the robot's 18650 Li-ion batteries with minimal effort (2 x 2800mAh 18650 Li-ion cells). The batteries should last about 1 year if you use the robot every day.
- 4x4 wheel drive - all wheel drive is a necessary design choice to support good locomotion across long hair carpets, floor bumps and clothing left on floor.
- Noise suppression - geared motors generated noise will be significantly reduced by 3 levels of sound insulation: geared motors rubber enclosure, rubber belt transmission to wheels and rubber tires.
- DIY customization and hacking SDK - our robots comprehensively broadcast their status using Android Intents to support integration with the 3rd party custom applications and hacks. So you can extend the robot's behavior for your own custom actions and integrate it with your apps.