In this paper we focus on the challenging problem of place categorization and semantic mapping a robot without environment-specific training. Motivated by their ongoing success in various visual recognition tasks, build our system upon state-of-the-art convolutional network. We overcome its closed-set limitations complementing network with series one-vs-all classifiers that can learn to recognize new classes online. Prior domain knowledge is incorporated embedding classification into...

We present Bayesian Controller Fusion (BCF): a hybrid control strategy that combines the strengths of traditional hand-crafted controllers and model-free deep reinforcement learning (RL). BCF thrives in robotics domain, where reliable but suboptimal priors exist for many tasks, RL from scratch remains unsafe data-inefficient. By fusing uncertainty-aware distributional outputs each system, arbitrates between them, exploiting their respective strengths. study on two real-world tasks involving...

In this paper we present for the first time a complete symbolic navigation system that performs goal-directed exploration to unfamiliar environments on physical robot. We introduce novel construct called abstract map link provided spatial information with observed and actual places in real world. Symbolic is using text recognition has been developed specifically application of reading door labels. study described paper, robot was floor plan destination. The destination specified by room...

Visually recognising a traversed route - regardless of whether seen during the day or night, in clear inclement conditions, summer winter is an important capability for navigating robots. Since SeqSLAM was introduced 2012, large body work has followed exploring how robotic systems can use algorithm to meet challenges posed by navigation changing environmental conditions. The following paper describes OpenSeqSLAM2.0, fully open-source toolbox visual place recognition under Beyond benefits...

Domestic and service robots have the potential to transform industries such as health care small-scale manufacturing, well homes in which we live. However, due overwhelming variety of tasks these will be expected complete, providing generic out-of-the-box solutions that meet needs every possible user is clearly intractable. To address this problem, must therefore not only capable learning how complete novel at run-time, but also informed by user. In letter demonstrate behaviour trees, a...

In this work we focus on improving the efficiency and generalisation of learned navigation strategies when transferred from its training environment to previously unseen ones. We present an extension residual reinforcement learning framework robotic manipulation literature adapt it vast unstructured environments that mobile robots can operate in. The concept is based a control effect add typical sub-optimal classical controller in order close performance gap, whilst guiding exploration...

This paper shows that by using only symbolic language phrases, a mobile robot can purposefully navigate to specified rooms in previously unexplored environments. The intelligently organises description of the unseen environment and "imagines" representative map, called abstract map. map is an internal representation topological structure spatial layout symbolically defined locations. To perform goal-directed exploration, creates high-level semantic plan reason about spaces beyond robot's...

Human navigation in built environments depends on symbolic spatial information which has unrealised potential to enhance robot capabilities. Information sources such as labels, signs, maps, planners, spoken directions, and navigational gestures communicate a wealth of the navigators environments; that robots typically ignore. We present system uses same employed by humans purposefully navigate unseen with level performance comparable humans. The novel data structure called abstract map...

Learning-based approaches often outperform hand-coded algorithmic solutions for many problems in robotics. However, learning long-horizon tasks on real robot hardware can be intractable, and transferring a learned policy from simulation to reality is still extremely challenging. We present novel approach model-free reinforcement that leverage existing sub-optimal as an prior during training deployment. During training, our gated fusion enables the guide initial stages of exploration,...

We introduce BenchBot, a novel software suite for benchmarking the performance of robotics research across both photorealistic 3D simulations and real robot platforms. BenchBot provides simple interface to sensorimotor capabilities when solving problems; an that is consistent regardless whether target platform simulated or robot. In this paper we outline system architecture, explore parallels between its user-centric design ideal development process devoid tangential engineering challenges....

Recent Semantic SLAM methods combine classical geometry-based estimation with deep learning-based object detection or semantic segmentation. In this paper we evaluate the quality of maps generated by state-of-the-art class- and instance-aware dense algorithms whose codes are publicly available explore impacts both segmentation pose have on maps. We obtain these results providing ground-truth and/or data from simulated environments. establish that is largest source error through our...

Visually recognising a traversed route - regardless of whether seen during the day or night, in clear inclement conditions, summer winter is an important capability for navigating robots. Since SeqSLAM was introduced 2012, large body work has followed exploring how robotic systems can use algorithm to meet challenges posed by navigation changing environmental conditions. The following paper describes OpenSeqSLAM2.0, fully open source toolbox visual place recognition under Beyond benefits...

We present a platform to foster research in active scene understanding, consisting of high-fidelity simulated environments and simple yet powerful API that controls mobile robot simulation reality. In contrast static, pre-recorded datasets focus on the perception aspect agency is top priority our work. provide three levels agency, allowing users control at varying difficulty realism. While most basic level provides pre-defined trajectories ground-truth localisation, more realistic allow us...