Autonomous race cars, such as in Formula Student Driverless, operate close to their physical handling limits. The resulting highly nonlinear vehicle behavior increases the path tracking complexity, especially on narrow tracks. Model Predictive Control (MPC) is commonly used to address this issue, a method whose performance is closely tied to the accuracy of the underlying prediction model. This paper presents a novel, real-time capable prediction model for autonomous race cars that adjusts to changing conditions by combining information from past runs and the current driving situation. Our model is divided into three consecutive submodels: a nominal Kinematic Bicycle Model, an offline Bayesian Linear Regression (BLR) model, and an online Sparse Gaussian Process Regression (SGPR) model. The proposed approach enables efficient integration of all available data without significantly increasing computational cost, ensuring high prediction accuracy and a quantitative uncertainty assessment right from the start of the run. Compared to existing approaches, an improvement in prediction accuracy of up to 57% was achieved. Further, we successfully demonstrated the practical applicability of the model within an MPC-based path tracking controller on a real Formula Student race car.

In autonomous racing, especially in competitions such as Formula Student Driverless, precise planning of the target velocity of a race car is crucial for competitive lap times and stable driving behavior. Especially at high speeds, Velocity Planning (VP) is a significant challenge as it has to be performed in real time, taking into account track layouts, environmental influences, mechanical tolerances, and the resulting control inaccuracies. In this paper, we present a novel approach to VP that dynamically adapts to such changing conditions. Instead of estimating the physical Tire-Road Friction Coefficient (TRFC), a continuous scaling factor is inferred indirectly from vehicle stability. This factor not only reflects the effective tire-road interaction but also captures effects of control inaccuracies. From this, we generate a continuous friction map, which serves as a robust, adaptive basis for computing the optimal target speed, accounting for both vehicle and environmental limits. Our proposed approach was evaluated on a real Formula Student race car, showing a lap time improvement of 35 % over ten laps and an average increase of 8 % compared to a non-adaptive approach.

Radio fingerprinting is a well-established concept for modeling the propagation of Wi-Fi, Bluetooth or other radio signals in a known environment for localization purposes. When using position estimates from such a model for trajectory tracking over time in a Bayesian framework such as a Particle Filter, it must be possible to evaluate arbitrary state hypotheses based on the fingerprinting model. Instead of having a single position estimate for current observations, the model target should be a continuous probability density function from which any hypothesis can be evaluated. This work proposes an interpolation approach for the position estimates of fingerprinting models. The interpolation is based on Gaussian Process Regression. The proposed approach is particularly capable of handling multimodalities and maintains high accuracy even in large environments with little calibration data. It is also efficient in terms of runtime, making it a suitable choice for real-time applications.

Smartphone-based indoor positioning and navigation remains a challenging task, as specialized technologies such as ultra-wideband (UWB) or Wi-Fi fine-time measurement are still niche and supported by only a few flagship smartphones. Therefore, standard technologies based on RSSI measurements, mainly Bluetooth Low Energy (BLE) and Wi-Fi, are used to obtain absolute positioning information of pedestrians inside buildings. Sensor fusion methods combine this with relative information from modeling human movement using sensor data provided by the smartphone’s IMU. It is also common practice to restrict this movement to the actual accessible areas of the building (e.g. restricting moving through walls), using spatial models based on the building’s floor plan. Without further assumptions, this complexity inevitably leads to a non-linear and non-Gaussian state space model. A common tool for (position) estimation in such scenarios is the broad class of particle filters. However, the use of such spatial constraints accelerates the well-known problem of sample impoverishment, which in the worst case can lead to the particle filter completely losing track, getting stuck and never recovering. This work begins with a brief presentation of an award-winning Indoor Positioning System (IPS) derived from previous work. Based on this, we present several approaches using Particle Distribution Optimization (PDO) that attempt to solve the impoverishment problem and ultimately lead to better overall positioning results. In the experiments, we compare them in two different buildings under realistic conditions and discuss the results in detail.

Acquiring accurate ground truth or fingerprint data for indoor localization typically demands specialized hardware and considerable effort. This paper introduces three post-processing methods to simplify the recording of such data within certain limitations. The methods can be categorized into fixed-interval smoothing where the complete recording data is available for the optimal state estimation and there is no live processing requirement. Our methods are based on a particle filter system, employing the CONDENSATION algorithm to estimate the hidden state. Particles represent positions and headings in 3D space. They are moved and evaluated using pedestrian dead reckoning based on IMU measurements and a map of the building only. The development and application of these methods is presented in this paper, demonstrating their effectiveness in reducing the effort required for accurate ground truth and fingerprint data recording.

This work presents SIMUL, a new dataset consisting of 550 minutes of annotated motion data from six synchronized IMUs placed consistently at the same strategically chosen positions on the bodies of 32 participants. With a focus on indoor localization, the selection fell on hand, feet, and trouser pockets. Due to the sensor’s synchronization, this selection allows, for example, to label the data recorded in the hand based on the data captured at the feet. For a better generalizability, the dataset was recorded freestyle under many different environmental conditions and walking speeds. Thus, indoors, numerous floor coverings such as stone, wood, carpet and PVC, as well as stairs are included in the data. The outdoor recordings additionally contain uneven surfaces such as paving stones and slopes. The annotation of the participant’s currently performed activity additionally allows this dataset to be used for activity recognition.

Radio frequency ranging protocols enable a device to estimate the distance to another device, based on signal propagation time. In theory, ranging protocols are promising for indoor localization as one can obtain the position of the pedestrian directly from the ranging results if the access point positions are known. However, in practice, indoor scenarios still pose a challenging problem as the observed distances vary greatly due to non-line-of-sight signal paths, delayed signal propagation, and general hardware inaccuracies. The IEEE 802.11-2016 (formerly IEEE 802.11mc) standard defines a RF ranging protocol for Wi-Fi, namely Fine Timing Measurement (FTM). In order to improve the position estimate, a novel sensor model for FTM is derived from observed data. It is shown that the FTM error varies with the actual distance to the access point. Within this work, different parameter sets are estimated from the observed data for skew normal distributions, depending on the actual distance. For these parameters, low-order polynomials are then fitted to obtain the distribution parameters as functions of the actual distance. Furthermore, a particle filter is described and evaluated in an industrial scenario using cheap Espressif ESP32-S2 IoT FTM access points in the 2.4 GHz band. The filter combines map information, Pedestrian Dead Reckoning, and our novel FTM sensor model to estimate the pedestrian's position in the building. Finally, the localization result of the particle filter is compared to another promising radio frequency ranging method: ultra-wideband.

Particle filters are a commonly used technique for sensor fusion in indoor localization use-cases. Multiple strategies exist, that control when the particle filter is updated and with which data. We take a look at three commonly used update strategies. The first runs a full particle filter update for every incoming measurement, the second uses a fixed-interval update rate and the third is triggered by events, such as detected steps. All of these strategies have different advantages and disadvantages. The first strategy, for example, has the problem that steps are recognized only after they have been completed - which makes for a constant temporal discrepancy between the proposal distribution and the measurements evaluated on top. Due to the configured delay, the fixed-interval strategy, in comparison, can pre-date incoming step events to mitigate this discrepancy. In this paper we present PIPF as a novel approach to combine advantages of the fixed-interval update strategy with advantages of the strategy that runs a full update per measurement. This works by using the transition to calculate a trajectory, which is then followed during the evaluation. The proposal distribution in the form of particles is interpolated on this trajectory for every incoming measurement, which removes the temporal discrepancy between both. To evaluate PIPF's performance and characteristics at different configurations, we compare it to a conventional fixed-interval particle filter on a real-world indoor positioning scenario.

It is common practice to use a sample-based representation to solve problems having a probabilistic interpretation. In many real world scenarios one is then interested in finding a best estimate of the underlying problem, e.g. the position of a robot. This is often done by means of simple parametric point estimators, providing the sample statistics. However, in complex scenarios this frequently results in a poor representation, due to multimodal densities and limited sample sizes. Recovering the probability density function using a kernel density estimation yields a promising approach to solve the state estimation problem i. e. finding the “real” most probable state, but comes with high computational costs. Especially in time critical and time sequential scenarios, this turns out to be impractical. Therefore, this work uses techniques from digital signal processing in the context of estimation theory, to allow rapid computations of kernel density estimates. The gains in computational efficiency are realized by substituting the Gaussian filter with an approximate filter based on the box filter. Our approach outperforms other state of the art solutions, due to a fully linear complexity and a negligible overhead, even for small sample sets. Finally, our findings are evaluated and tested within a real world sensor fusion system.

In recent research, indoor localisation systems are often based upon a recursive state estimation using particle filtering. Within this context, sample impoverishment is a crucial problem causing the position estimation to lose track or get stuck within a demarcated area. The sample impoverishment problem can therefore be described as a too small particle diversity, unable to sample enough particles into proper regions of the dynamic system. Restrictive transition models, as they are used in indoor localisation, also enhance this effect significantly. However, an accurate position estimation requires a certain degree of focus and thus behaves contrary to the need of diversity. Therefore we propose a new method that is able to deal with the trade-off between the need of diversity and focus by deploying an interacting multiple model particle filter (IMMPF) for jump Markov non-linear systems. We combine two similar particle filters using a non-trivial Markov switching process, depending upon the Kullback-Leibler divergence and a Wi-Fi quality factor. The main benefit of this approach is an easy adaptation to other localisation approaches based on particle filters.

Indoor localisation continues to be a topic of growing importance. Despite the advances made, several profound problems are still present. For example, estimating an accurate position from a multimodal distribution or recovering from the influence of faulty measurements. Within this work, we solve such problems with the help of Monte Carlo smoothing methods, namely forward-backward smoother and backward simulation. In contrast to normal filtering procedures like particle filtering, smoothing methods are able to incorporate future measurements instead of just using current and past data. This enables many possibilities for further improving the position estimation. Both smoothing techniques are deployed as fixed-lag and fixed-interval smoother and a novel approach for incorporating them easily within a conventional localisation system is presented. All this is evaluated on four floors within our faculty building. The results show that smoothing methods offer a great tool for improving the overall localisation. Especially fixed-lag smoothing provides a great runtime support by reducing temporal errors and improving the overall estimation with affordable costs.

We present an indoor localisation system that integrates different sensor modalities, namely Wi-Fi, barometer, iBeacons, step-detection and turn-detection for localisation of pedestrians within buildings over multiple floors. To model the pedestrian's movement, which is constrained by walls and other obstacles, we propose a state transition based upon random walks on graphs. This model also frees us from the burden of frequently updating the system. In addition we make use of barometer information to estimate the current floor. Furthermore, we present a statistical approach to avoid the incorporation of faulty heading information caused by changing the smartphone's position. The evaluation of the system within a 77m × 55m sized building with 4 floors shows that high accuracy can be achieved while also keeping the update-rates low.

Indoor positioning systems are required for many new applications and, ideally, should provide high accuracy at zero costs for initial setup, maintenance and per user. Many approaches thus use an existing Wi-Fi infrastructure and smart-phones for pedestrian location estimation. While the well-known Wi-Fi fingerprinting provides good localization accuracy down to one meter, necessary time and costs are tremendous. Alternatives, like model-based signal strength estimation, are easy to setup but supply viable results only for line of sight conditions. To provide an inexpensive yet accurate solution, we combine Wi-Fi localization using a signal strength prediction model together with step/turn-detection based on a smartphone's accelerometer/gyroscope and incorporate a priori knowledge utilizing the building's floorplan. Our technique uses a statistical model for both, Wi-Fi and step/turn-detection, to calculate the probability of the pedestrian residing at some arbitrary position and leverages the building's floorplan to determine the likelihood for any possible movement between two positions. Latter probabilities are combined with the two densities from Wi-Fi and step/turn-detection applying recursive density estimation implemented using well-known particle filtering techniques. While the density estimated through Wi-Fi measurements provides a vague, absolute position at a significant uncertainty, the step-detector supplies a fine resolution at the downside of a cumulative error due to its estimation relative to previous steps. The fusion of these two sensor densities compensates for this error and the floorplan further enhances the estimation result. We will show that our statistical approach provides a robust, long-term stable location estimation, much better than Wi-Fi on its own while requiring only a few (empirical) parameters.

This paper presents a method for indoor localization of humans. Our new approach combines an imaged-based position estimation with a given step and turn detection. Estimating the position of an object is not only a question of accuracy, it is also a question of performance and time required. Our approach uses a fusion of an uncalibrated depth-sensor with a smartphone's accelerometer and gyroscope. Both components do not rely on time-consuming practices like fingerprinting and calibration techniques. This approach allows for a real time-tracking by using well-known methods of recursive density propagation and particle filtering. Unlike other image-based methods in autonomous robotics, the depth sensors are mounted at a fixed position. Therefore we will show how our new statistical sensor model covers the four main conditions of an image-based approach: a person is either inside or outside the field of view and is detected or not detected by the depth sensor. This involves the possibility to estimate the position of multiple persons at each point in time. Finally, the experimental results show how the integration of an image-based approach increases the accuracy of the localization estimation and counteracts the increasing error of the given step and turn detection.

Object recognition problems in computer vision are often based on single image data processing. In various applications this processing can be extended to a complete sequence of images, usually received passively. In contrast, we propose a method for active object recognition, where a camera is selectively moved around a considered object. Doing so, we aim at reliable classification results with a clearly reduced amount of necessary views by optimizing the camera movement for the access of new viewpoints (viewpoint selection). Therefore, the optimization criterion is the gain of class discriminative information when observing the appropriate next image. We show how to apply an unsupervised reinforcement learning algorithm to that problem. Specifically, we focus on the modeling of continuous states, continuous actions and supporting rewards for an optimized recognition. We also present an algorithm for the sequential fusion of gathered image information and we combine all these components into a single framework. The experimental evaluations are split into results for synthetic and real objects with one- or two-dimensional camera actions, respectively. This allows the systematic evaluation of the theoretical correctness as well as the practical applicability of the proposed method. Our experiments showed that the proposed combined viewpoint selection and viewpoint fusion approach is able to significantly improve the recognition rates compared to passive object recognition with randomly chosen views.

In the past decades, most object recognition systems were based on passive approaches. But in the last few years a lot of research was done in the field of active object recognition, that is selectively moving a sensor/camera around a considered object in order to acquire as much information about it as possible. In this paper we present an active object recognition approach that solves the problem of choosing optimal views (viewpoint selection) and iteratively fuses the gained information for an optimal 3D object recognition (viewpoint fusion) in an integrated manner. Therefore, we apply a method for the fusion of multiple views with respect to the knowledge about the assumed camera movement between them. For viewpoint selection we formally define the choice of additional views as an optimization problem. We show how to use reinforcement learning for this purpose and perform a training without user interaction. In this context we focus on the modeling of continuous states, continuous, one-dimensional actions and supporting rewards for an optimized recognition of real objects. The experimental results show that our combined viewpoint selection and viewpoint fusion approach is able to significantly improve the recognition rates compared to passive object recognition with randomly chosen views. 1

In speech technology more and more databases of spoken language are becoming available. For research the availability of these data offers the possibility to study huge corpora. Apart from the fact that these corpora may be represented in different formats, it is sometimes difficult to relate annotations of one corpus to those of another corpus. This contribution argues for a representation of information in speech corpora that allows for the integrated representation of information on various levels of description in XML. Secondly, the study of huge amounts of speech data requires adequate retrieval mechanisms. A query architecture is described that allows for the retrieval of encoded entities by specifying their properties or various relations to other entities. The output of the query processor is represented in XML and thus can be used for further queries or a new level of description. The work presented here is part of the results of the MATE project (http://mate.mip.ou.dk).

In this paper we present an innovative approach to speech understanding which is based on a fine-grained knowledge representation automatically compiled from a semantic network and on iterative optimization. Besides allowing an efficient exploitation of parallelism, any-time capability is provided since after each iteration step a (sub-)optimal solution is always available. We apply this approach to a real--world task, which is a dialog system able to answer queries about the German train timetable. In order to speed up the search for the best interpretation of an utterance we make use of statistical methods, e.g. neural networks, n-grams, and classification trees, which are trained on application relevant utterances collected over the public telephone network. At the moment the real--time factor for interpreting the initial user's utterance is 0.7.