From vision to actions: Towards adaptive and autonomous humanoid robots

Although robotics research has seen advances over the last decades robots are still not in widespread use outside industrial applications. Yet a range of proposed scenarios have robots working together, helping and coexisting with humans in daily life. In all these a clear need to deal with a more unstructured, changing environment arises. I herein present a system that aims to overcome the limitations of highly complex robotic systems, in terms of autonomy and adaptation. The main focus of research is to investigate the use of visual feedback for improving reaching and grasping capabilities of complex robots. To facilitate this a combined integration of computer vision and machine learning techniques is employed. From a robot vision point of view the combination of domain knowledge from both imaging processing and machine learning techniques, can expand the capabilities of robots. I present a novel framework called Cartesian Genetic Programming for Image Processing (CGP-IP). CGP-IP can be trained to detect objects in the incoming camera streams and successfully demonstrated on many different problem domains. The approach requires only a few training images (it was tested with 5 to 10 images per experiment) is fast, scalable and robust yet requires very small training sets. Additionally, it can generate human readable programs that can be further customized and tuned. While CGP-IP is a supervised-learning technique, I show an integration on the iCub, that allows for the autonomous learning of object detection and identification. Finally this dissertation includes two proof-of-concepts that integrate the motion and action sides. First, reactive reaching and grasping is shown. It allows the robot to avoid obstacles detected in the visual stream, while reaching for the intended target object. Furthermore the integration enables us to use the robot in non-static environments, i.e. the reaching is adapted on-the- fly from the visual feedback received, e.g. when an obstacle is moved into the trajectory. The second integration highlights the capabilities of these frameworks, by improving the visual detection by performing object manipulation actions.

Adding a receding horizon to Locally Weighted Regression for learning robot control

There have been notable advances in learning to control complex robotic systems using methods such as Locally Weighted Regression (LWR). In this paper we explore some potential limits of LWR for robotic applications, particularly investigating its application to systems with a long horizon of temporal dependence. We define the horizon of temporal dependence as the delay from a control input to a desired change in output. LWR alone cannot be used in a temporally dependent system to find meaningful control values from only the current state variables and output, as the relationship between the input and the current state is under-constrained. By introducing a receding horizon of the future output states of the system, we show that sufficient constraint is applied to learn good solutions through LWR. The new method, Receding Horizon Locally Weighted Regression (RH-LWR), is demonstrated through one-shot learning on a real Series Elastic Actuator controlling a pendulum.

Learning multidimensional joint control of a robot using receding horizon locally weighted regression

In this paper we explore the ability of a recent model-based learning technique Receding Horizon Locally Weighted Regression (RH-LWR) useful for learning temporally dependent systems. In particular this paper investigates the application of RH-LWR to learn control of Multiple-input Multiple-output robot systems. RH-LWR is demonstrated through learning joint velocity and position control of a three Degree of Freedom (DoF) rigid body robot.

Locally weighted learning methods for non-rigid robot control

This thesis develops a novel approach to robot control that learns to account for a robot's dynamic complexities while executing various control tasks using inspiration from biological sensorimotor control and machine learning. A robot that can learn its own control system can account for complex situations and adapt to changes in control conditions to maximise its performance and reliability in the real world. This research has developed two novel learning methods, with the aim of solving issues with learning control of non-rigid robots that incorporate additional dynamic complexities. The new learning control system was evaluated on a real three degree-of-freedom elastic joint robot arm with a number of experiments: initially validating the learning method and testing its ability to generalise to new tasks, then evaluating the system during a learning control task requiring continuous online model adaptation.

Learning spatial concepts from RatSLAM representations

RatSLAM is a biologically-inspired visual SLAM and navigation system that has been shown to be effective indoors and outdoors on real robots. The spatial representation at the core of RatSLAM, the experience map, forms in a distributed fashion as the robot learns the environment. The activity in RatSLAM’s experience map possesses some geometric properties, but still does not represent the world in a human readable form. A new system, dubbed RatChat, has been introduced to enable meaningful communication with the robot. The intention is to use the “language games” paradigm to build spatial concepts that can be used as the basis for communication. This paper describes the first step in the language game experiments, showing the potential for meaningful categorization of the spatial representations in RatSLAM.

Representation and learning of visual information for pose recognition

Recovering position from sensor information is an important problem in mobile robotics, known as localisation. Localisation requires a map or some other description of the environment to provide the robot with a context to interpret sensor data. The mobile robot system under discussion is using an artificial neural representation of position. Building a geometrical map of the environment with a single camera and artificial neural networks is difficult. Instead it would be simpler to learn position as a function of the visual input. Usually when learning images, an intermediate representation is employed. An appropriate starting point for biologically plausible image representation is the complex cells of the visual cortex, which have invariance properties that appear useful for localisation. The effectiveness for localisation of two different complex cell models are evaluated. Finally the ability of a simple neural network with single shot learning to recognise these representations and localise a robot is examined.

Learning to avoid indoor obstacles from optical flow

Optical flow (OF) is a powerful motion cue that captures the fusion of two important properties for the task of obstacle avoidance − 3D self-motion and 3D environmental surroundings. The problem of extracting such information for obstacle avoidance is commonly addressed through quantitative techniques such as time-to-contact and divergence, which are highly sensitive to noise in the OF image. This paper presents a new strategy towards obstacle avoidance in an indoor setting, using the combination of quantitative and structural properties of the OF field, coupled with the flexibility and efficiency of a machine learning system.The resulting system is able to effectively control the robot in real-time, avoiding obstacles in familiar and unfamiliar indoor environments, under given motion constraints. Furthermore, through the examination of the networks internal weights, we show how OF properties are being used toward the detection of these indoor obstacles.

Biologically inspired joint control for a humanoid robot

The GuRoo is a 1.2 m tall, 23 degree of freedom humanoid constructed at the University of Queensland for research into humanoid robotics. The key challenge being addressed by the GuRoo project is the development of appropriate learning strategies for control and coordination of the robot's many joints. The development of learning strategies is seen as a way to side-step the inherent intricacy of modeling a multi-DOF biped robot. This paper outlines the approach taken to generate an appropriate control scheme for the joints of the GuRoo. The paper demonstrates the determination of local feedback control parameters using a genetic algorithm. The feedback loop is then augmented by a predictive modulator that learns a form of feed-forward control to overcome the irregular loads experienced at each joint during the gait cycle. The predictive modulator is based on the CMAC architecture. Results from tests on the GuRoo platform show that both systems provide improvements in stability and tracking of joint control.

Cerebellar augmented joint control for a humanoid robot

The joints of a humanoid robot experience disturbances of markedly different magnitudes during the course of a walking gait. Consequently, simple feedback control techniques poorly track desired joint trajectories. This paper explores the addition of a control system inspired by the architecture of the cerebellum to improve system response. This system learns to compensate the changes in load that occur during a cycle of motion. The joint compensation scheme, called Trajectory Error Learning, augments the existing feedback control loop on a humanoid robot. The results from tests on the GuRoo platform show an improvement in system response for the system when augmented with the cerebellar compensator.

Hybrid robot control and SLAM for persistent navigation and mapping

For a mobile robot to operate autonomously in real-world environments, it must have an effective control system and a navigation system capable of providing robust localization, path planning and path execution. In this paper we describe the work investigating synergies between mapping and control systems. We have integrated development of a control system for navigating mobile robots and a robot SLAM system. The control system is hybrid in nature and tightly coupled with the SLAM system; it uses a combination of high and low level deliberative and reactive control processes to perform obstacle avoidance, exploration, global navigation and recharging, and draws upon the map learning and localization capabilities of the SLAM system. The effectiveness of this hybrid, multi-level approach was evaluated in the context of a delivery robot scenario. Over a period of two weeks the robot performed 1143 delivery tasks to 11 different locations with only one delivery failure (from which it recovered), travelled a total distance of more than 40km, and recharged autonomously a total of 23 times. In this paper we describe the combined control and SLAM system and discuss insights gained from its successful application in a real-world context.

Applying incremental EM to Bayesian classifiers in the learning of hyperspectral remote sensing data

In this paper, we apply the incremental EM method to Bayesian Network Classifiers to learn and interpret hyperspectral sensor data in robotic planetary missions. Hyperspectral image spectroscopy is an emerging technique for geological investigations from airborne or orbital sensors. Many spacecraft carry spectroscopic equipment as wavelengths outside the visible light in the electromagnetic spectrum give much greater information about an object. The algorithm used is an extension to the standard Expectation Maximisation (EM). The incremental method allows us to learn and interpret the data as they become available. Two Bayesian network classifiers were tested: the Naive Bayes, and the Tree-Augmented-Naive Bayes structures. Our preliminary experiments show that incremental learning with unlabelled data can improve the accuracy of the classifier.

Learning informative features for indoor traversability

This paper presents a method for automatic terrain classification, using a cheap monocular camera in conjunction with a robot’s stall sensor. A first step is to have the robot generate a training set of labelled images. Several techniques are then evaluated for preprocessing the images, reducing their dimensionality, and building a classifier. Finally, the classifier is implemented and used online by an indoor robot. Results are presented, demonstrating an increased level of autonomy.

STALKERBOT : learning to navigate dynamic human environments by following people

Service robots that operate in human environments will accomplish tasks most efficiently and least disruptively if they have the capability to mimic and understand the motion patterns of the people in their workspace. This work demonstrates how a robot can create a humancentric navigational map online, and that this map re ects changes in the environment that trigger altered motion patterns of people. An RGBD sensor mounted on the robot is used to detect and track people moving through the environment. The trajectories are clustered online and organised into a tree-like probabilistic data structure which can be used to detect anomalous trajectories. A costmap is reverse engineered from the clustered trajectories that can then inform the robot's onboard planning process. Results show that the resultant paths taken by the robot mimic expected human behaviour and can allow the robot to respond to altered human motion behaviours in the environment.