Inhalt des Dokuments
This page lists software, web services and tutorials that are developed and maintained as part of our research at RBO Lab, and which are publicly available for private use and research.
Online Interactive Perception
Our code of our perceptual system for Online Interactive Perception of articulated objects. It includes our methods presented in at IROS14 (PDF, 1,8 MB)  and ICRA16 (PDF, 2,3 MB) . The system extracts patterns of motion at different levels (point feature motion, rigid body motion, kinematic structure motion) and infers the kinematic structure and state of the interacted articulated objects. It uses an RGB-D stream of an interaction as input. Optionally, it can reconstruct the shape of the moving parts and use it to improve tracking. More info at omip wiki .
Repository: https://github.com/tu-rbo/omip  and https://github.com/tu-rbo/omip_msgs 
Accompanies papers: Online Interactive Perception of Articulated Objects with Multi-Level Recursive Estimation Based on Task-Specific Priors (PDF, 1,8 MB)  and An Integrated Approach to Visual Perception of Articulated Objects (PDF, 2,3 MB) 
Maintainers: Roberto Martín-Martín 
Project: Interactive Perception 
Learning State Representations with Robotic Priors
This simple python script contains the essence of our work on learning state representations with robotic priors and complements the paper to provide sufficient detail for reproducing our results and for reusing the method in other research while minimizing code overhead.
Repository: github.com/tu-rbo/learning-state-representations-with-robotic-priors 
Accompanies paper: Learning State Representations with Robotic Priors 
Maintainers: Rico Jonschkowski 
Project: Representation Learning 
Probabilistic Multi-Class Segmentation for the Amazon Picking Challenge
This repository includes the code and data for the object perception method of our winning entry to the 2015 Amazon Picking Challenge. The object segmentation method and the analysis from the accompaning paper are implemented in python. The data can be used to develop and benchmark new perception methods in this setting.
Repository: gitlab.tubit.tu-berlin.de/rbo-lab/rbo-apc-object-segmentation 
Accompanies paper: Probabilistic Multi-Class Segmentation for the Amazon Picking Challenge 
Maintainers: Rico Jonschkowski 
Project: Amazon Picking Challenge 2015 
concarne: Learning with Side Information
concarne is a lightweight python framework for learning with side information (aka privileged information). concarne implements a variety of different patterns that enable to apply side information. As it builds upon Theano  and lasagne , you can use neural network structures that you have developed yourself and easily combine them with the side information learning task.
Repository: https://github.com/tu-rbo/concarne 
Accompanies paper: Patterns for Learning with Side Information 
Maintainers: Sebastian Höfer, Rico Jonschkowski 
Research project: Representation Learning 
RBO Aleph: A Web Service for Automated Protein Structure Prediction
RBO Aleph provides a fully automatic protein structure prediction pipeline. The focus of our system is ab initio structure prediction.
RBO Aleph has been evaluated to be one of the leading ab initio structure prediction methods in CASP11.
A web interface of to RBO Aleph is available here:
Contact: Mahmoud Mabrouk 
EPC-map: Combining Evolutionary and Physicochemical Information for Contact Prediction
Our approach integrates decoy-based contact prediction with machine learning and evolutionary information from multiple-sequence alignments, called EPC-map (using Evolutionary and Physicochemical information to predict Contact maps).
This method ranked 2nd for long+medium range contacts and 5th for long-range contacts in CASP11.
A web service implementing this method is available here:
Contact: Kolja Stahl 
Tutorial on building PneuFlex actuators
PneuFlex actuators are a version of so called soft pneumatic continuum actuators. Their production process is relatively simple, and can accomodate complex actuator shapes. This tutorial  explains how to build them.
Contact: Raphael Deimel, Vincent Wall 
Pneumaticbox Control System
The Pneumaticbox is a specific hardware combined with an embedded software system to provide a complete control solution for PneuFlex actuators:
- The Hardware  consists of valves, pressure sensors, common I/O interfaces (USB, SPI, I2C), and an embedded computer.
- The pneumaticbox-airserver  runs on the Pneumaticbox to abstract hardware I/O and runs real-time controllers on behalf of networked clients. It includes controllers for straight forward control of valve opening times , but also more advanced controllers for air mass and air mass flow  control. Additionally, safety measures in the form of a pressure limiter  and a pressure watchdog (emergency stop)  are implemented.
- The python-pneumaticbox  is the Python client library. It can be used for easy integration into existing systems or direct scripting of experiments.
- Raphael Deimel and Oliver Brock. Soft Hands for Reliable Grasping Strategies . Soft Robotics, Soft Hands, Grasping. Springer-Verlag, chap. Soft Hands for Reliable Grasping Strategies, 211–221, 2015.
- Raphael Deimel and Oliver Brock. A Novel Type of Compliant and Underactuated Robotic Hand for Dexterous Grasping . The International Journal of Robotics Research 35(1-3):161-185, 2016.
- Raphael Deimel and Marcel Radke and Oliver Brock. Mass Control of Pneumatic Soft Continuum Actuators with Commodity Components . IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 774–779, 2016.
Contact: Raphael Deimel, Vincent Wall