We introduce a Generalizable Neural Radiance Field approach for predicting 3D workspace occupancy from egocentric robot observations. Unlike prior methods operating in camera-centric coordinates, our model constructs occupancy representations in a global workspace frame, making it directly applicable to robotic manipulation. The model integrates flexible source views and generalizes to unseen object arrangements without scene-specific finetuning. We demonstrate the approach on a humanoid robot and evaluate predicted geometry against 3D sensor ground truth. Trained on 40 real scenes, our model achieves 26mm reconstruction error, including occluded regions, validating its ability to infer complete 3D occupancy beyond traditional stereo vision methods.

Reinforcement learning (RL) is a powerful technique for training intelligent agents, but understanding why these agents make specific decisions can be quite challenging. This lack of transparency in RL models has been a long-standing problem, making it difficult for users to grasp the reasons behind an agent's behaviour. Various approaches have been explored to address this problem, with one promising avenue being reward decomposition (RD). RD is appealing as it sidesteps some of the concerns associated with other methods that attempt to rationalize an agent's behaviour in a post-hoc manner. RD works by exposing various facets of the rewards that contribute to the agent's objectives during training. However, RD alone has limitations as it primarily offers insights based on sub-rewards and does not delve into the intricate cause-and-effect relationships that occur within an RL agent's neural model. In this paper, we present an extension of RD that goes beyond sub-rewards to provide more informative explanations. Our approach is centred on a causal learning framework that leverages information-theoretic measures for explanation objectives that encourage three crucial properties of causal factors: causal sufficiency, sparseness, and orthogonality. These properties help us distill the cause-and-effect relationships between the agent's states and actions or rewards, allowing for a deeper understanding of its decision-making processes. Our framework is designed to generate local explanations and can be applied to a wide range of RL tasks with multiple reward channels. Through a series of experiments, we demonstrate that our approach offers more meaningful and insightful explanations for the agent's action selections.

Explaining the behaviour of intelligent agents learned by reinforcement learning (RL) to humans is challenging yet crucial due to their incomprehensible proprioceptive states, variational intermediate goals, and resultant unpredictability. Moreover, one-step explanations for RL agents can be ambiguous as they fail to account for the agent's future behaviour at each transition, adding to the complexity of explaining robot actions. By leveraging abstracted actions that map to task-specific primitives, we avoid explanations on the movement level. To further improve the transparency and explainability of robotic systems, we propose an explainable Q-Map learning framework that combines reward decomposition (RD) with abstracted action spaces, allowing for non-ambiguous and high-level explanations based on object properties in the task. We demonstrate the effectiveness of our framework through quantitative and qualitative analysis of two robotic scenarios, showcasing visual and textual explanations, from output artefacts of RD explanations, that are easy for humans to comprehend. Additionally, we demonstrate the versatility of integrating these artefacts with large language models (LLMs) for reasoning and interactive querying.

Neural fields are neural networks which map coordinates to a desired signal. When a neural field should jointly model multiple signals, and not memorize only one, it needs to be conditioned on a latent code which describes the signal at hand. Despite being an important aspect, there has been little research on conditioning strategies for neural fields. In this work, we explore the use of neural fields as decoders for 2D semantic segmentation. For this task, we compare three conditioning methods, simple concatenation of the latent code, Feature Wise Linear Modulation (FiLM), and Cross-Attention, in conjunction with latent codes which either describe the full image or only a local region of the image. Our results show a considerable difference in performance between the examined conditioning strategies. Furthermore, we show that conditioning via Cross-Attention achieves the best results and is competitive with a CNN-based decoder for semantic segmentation.

Recent machine learning advances demonstrate the effectiveness of zero-shot models trained on large amounts of data collected from the internet. Among these, CLIP (Contrastive Language-Image Pre-training) has been introduced as a multimodal model with high accuracy on a number of different tasks and domains. However, the unconstrained nature of the model begs the question whether it can be deployed in open-domain real-word applications effectively in front of non-technical users. In this paper, we evaluate whether CLIP can be used for multimodal video retrieval in a real-world environment. For this purpose, we implemented impa , an efficient shot-based retrieval system powered by CLIP. We additionally implemented advanced query functionality in a unified graphical user interface to facilitate an intuitive and efficient usage of CLIP for video retrieval tasks. Finally, we empirically evaluated our retrieval system by performing a user study with video editing professionals and journalists working in the TV news media industry. After having the participants solve open-domain video retrieval tasks, we collected data via questionnaires, interviews, and UI interaction logs. Our evaluation focused on the perceived intuitiveness of retrieval using natural language, retrieval accuracy, and how users interacted with the system’s UI. We found that our advanced features yield higher task accuracy, user ratings, and more efficient queries. Overall, our results show the importance of designing intuitive and efficient user interfaces to be able to deploy large models such as CLIP effectively in real-world scenarios.

One-step reinforcement learning explanation methods account for individual actions but fail to consider the agent’s future behavior, which can make their interpretation ambiguous. We propose to address this limitation by providing hierarchical goals as context for one-step explanations. By considering the current hierarchical goal as a context, one-step explanations can be interpreted with higher certainty, as the agent’s future behavior is more predictable. We combine reward decomposition with hierarchical reinforcement learning into a novel explainable reinforcement learning framework, which yields more interpretable, goal-contextualized one-step explanations. With a qualitative analysis of one-step reward decomposition explanations, we first show that their interpretability is indeed limited in scenarios with multiple, different optimal policies—a characteristic shared by other one-step explanation methods. Then, we show that our framework retains high interpretability in such cases, as the hierarchical goal can be considered as context for the explanation. To the best of our knowledge, our work is the first to investigate hierarchical goals not as an explanation directly but as additional context for one-step reinforcement learning explanations.

Recognizing emotions from text in multimodal architectures has yielded promising results, surpassing video and audio modalities under certain circumstances. However, the method by which multimodal data is collected can be significant for recognizing emotional features in language. In this paper, we address the influence data collection methodology has on two multimodal emotion recognition datasets, the IEMOCAP dataset and the OMG-Emotion Behavior dataset, by analyzing textual dataset compositions and emotion recognition accuracy. Experiments with the full IEMOCAP dataset indicate that the composition negatively influences generalization performance when compared to the OMG-Emotion Behavior dataset. We conclude by discussing the impact this may have on HRI experiments.

In domains where computational resources and labeled data are limited, such as in robotics, deep networks with millions of weights might not be the optimal solution. In this paper, we introduce a connectivity scheme for pyramidal architectures to increase their capacity for learning features. Experiments on facial expression recognition of unseen people demonstrate that our approach is a potential candidate for applications with restricted resources, due to good generalization performance and low computational cost. We show that our approach generalizes as well as convolutional architectures in this task but uses fewer trainable parameters and is more robust for low-resolution faces.

Ensemble methods, traditionally built with independently trained de-correlated models, have proven to be efficient methods for reducing the remaining residual generalization error, which results in robust and accurate methods for real-world applications. In the context of deep learning, however, training an ensemble of deep networks is costly and generates high redundancy which is inefficient. In this paper, we present experiments on Ensembles with Shared Representations (ESRs) based on convolutional networks to demonstrate, quantitatively and qualitatively, their data processing efficiency and scalability to large-scale datasets of facial expressions. We show that redundancy and computational load can be dramatically reduced by varying the branching level of the ESR without loss of diversity and generalization power, which are both important for ensemble performance. Experiments on large-scale datasets suggest that ESRs reduce the remaining residual generalization error on the AffectNet and FER+ datasets, reach human-level performance, and outperform state-of-the-art methods on facial expression recognition in the wild using emotion and affect concepts.

The recognition of emotion and dialogue acts enriches conversational analysis and help to build natural dialogue systems. Emotion interpretation makes us understand feelings and dialogue acts reflect the intentions and performative functions in the utterances. However, most of the textual and multi-modal conversational emotion corpora contain only emotion labels but not dialogue acts. To address this problem, we propose to use a pool of various recurrent neural models trained on a dialogue act corpus, with and without context. These neural models annotate the emotion corpora with dialogue act labels, and an ensemble annotator extracts the final dialogue act label. We annotated two accessible multi-modal emotion corpora: IEMOCAP and MELD. We analyzed the co-occurrence of emotion and dialogue act labels and discovered specific relations. For example, Accept/Agree dialogue acts often occur with the Joy emotion, Apology with Sadness, and Thanking with Joy. We make the Emotional Dialogue Acts (EDA) corpus publicly available to the research community for further study and analysis.

Recent studies have revealed the key importance of modelling personality in robots to improve interaction quality by empowering them with social-intelligence capabilities. Most research relies on verbal and non-verbal features related to personality traits that are highly context-dependent. Hence, analysing how humans behave in a given context is crucial to evaluate which of those social cues are effective. For this purpose, we designed an assistive memory game, in which participants were asked to play the game obtaining support from an introvert or extroverted helper, whether from a human or robot. In this context, we aim to (i) explore whether selective verbal and non-verbal social cues related to personality can be modelled in a robot, (ii) evaluate the efficiency of a statistical decision-making algorithm employed by the robot to provide adaptive assistance, and (iii) assess the validity of the similarity attraction principle. Specifically, we conducted two user studies. In the human–human study (N=31), we explored the effects of helper’s personality on participants’ performance and extracted distinctive verbal and non-verbal social cues from the human helper. In the human–robot study (N=24), we modelled the extracted social cues in the robot and evaluated its effectiveness on participants’ performance. Our findings showed that participants were able to distinguish between robots’ personalities, and not between the level of autonomy of the robot (Wizard-of-Oz vs fully autonomous). Finally, we found that participants achieved better performance with a robot helper that had a similar personality to them, or a human helper that had a different personality.

Sentiment analysis has transitioned from classifying the sentiment of an entire sentence to providing the contextual information of what targets exist in a sentence, what sentiment the individual targets have, and what the causal words responsible for that sentiment are. However, this has led to elaborate requirements being placed on the datasets needed to train neural networks on the joint triplet task of determining an entity, its sentiment, and the causal words for that sentiment. Requiring this kind of data for training systems is problematic, as they suffer from stacking subjective annotations and domain over-fitting leading to poor model generalisation when applied in new contexts. These problems are also likely to be compounded as we attempt to jointly determine additional contextual elements in the future. To mitigate these problems, we present a hybrid neural-symbolic method utilising a Dependency Tree-LSTM's compositional sentiment parse structure and complementary symbolic rules to correctly extract target-sentiment-cause triplets from sentences without the need for triplet training data. We show that this method has the potential to perform in line with state-of-the-art approaches while also simplifying the data required and providing a degree of interpretability through the Tree-LSTM.

Adversarial examples are artificially modified input samples which lead to misclassifications, while not being detectable by humans. These adversarial examples are a challenge for many tasks such as image and text classification, especially as research shows that many adversarial examples are transferable between different classifiers. In this work, we evaluate the performance of a popular defensive strategy for adversarial examples called defensive distillation, which can be successful in hardening neural networks against adversarial examples in the image domain. However, instead of applying defensive distillation to networks for image classification, we examine, for the first time, its performance on text classification tasks and also evaluate its effect on the transferability of adversarial text examples. Our results indicate that defensive distillation only has a minimal impact on text classifying neural networks and does neither help with increasing their robustness against adversarial examples nor prevent the transferability of adversarial examples between neural networks.

The complexity of bipedal locomotion may be attributed to the difficulty in synchronizing joint movements while at the same time achieving high-level objectives such as walking in a particular direction. Artificial central pattern generators (CPGs) can produce synchronized joint movements and have been used in the past for bipedal locomotion. However, most existing CPG-based approaches do not address the problem of high-level control explicitly. We propose a novel hierarchical control mechanism for bipedal locomotion where an optimized CPG network is used for joint control and a neural network acts as a high-level controller for modulating the CPG network. By separating motion generation from motion modulation, the high-level controller does not need to control individual joints directly but instead can develop to achieve a higher goal using a low-dimensional control signal. The feasibility of the hierarchical controller is demonstrated through simulation experiments using the Neuro-Inspired Companion (NICO) robot. Experimental results demonstrate the controller's ability to function even without the availability of an exact robot model.

Explaining the outcome of deep learning decisions based on affect is challenging but necessary if we expect social companion robots to interact with users on an emotional level. In this paper, we present a commonsense approach that utilizes an interpretable hybrid neural-symbolic system to associate extracted targets, noun chunks determined to be associated with the expressed emotion, with affective labels from a natural language expression. We leverage a pre-trained neural network that is well adapted to tree and sub-tree processing, the Dependency Tree-LSTM, to learn the affect labels of dynamic targets, determined through symbolic rules, in natural language. We find that making use of the unique properties of the recursive network provides higher accuracy and interpretability when compared to other unstructured and sequential methods for determining target-affect associations in an aspect-based sentiment analysis task.

Deep reinforcement learning has recently gained a focus on problems where policy or value functions are based on universal value function approximators (UVFAs) which renders them independent of goals. Evidence exists that the sampling of goals has a strong effect on the learning performance, and the problem of optimizing the goal sampling is frequently tackled with intrinsic motivation methods. However, there is a lack of general mechanisms that focus on goal sampling in the context of deep reinforcement learning based on UVFAs. In this work, we introduce goal masking as a method to estimate a goal's difficulty level and to exploit this estimation to realize curriculum learning. Our results indicate that focusing on goals with a medium difficulty level is appropriate for deep deterministic policy gradient (DDPG) methods, while an “aim for the stars and reach the moon-strategy”, where difficult goals are sampled much more often than simple goals, leads to the best learning performance in cases where DDPG is combined with hindsight experience replay (HER).

The recognition of emotion and dialogue acts enriches conversational analysis and help to build natural dialogue systems. Emotion interpretation makes us understand feelings and dialogue acts reflect the intentions and performative functions in the utterances. However, most of the textual and multi-modal conversational emotion corpora contain only emotion labels but not dialogue acts. To address this problem, we propose to use a pool of various recurrent neural models trained on a dialogue act corpus, with and without context. These neural models annotate the emotion corpora with dialogue act labels, and an ensemble annotator extracts the final dialogue act label. We annotated two accessible multi-modal emotion corpora: IEMOCAP and MELD. We analyzed the co-occurrence of emotion and dialogue act labels and discovered specific relations. For example, Accept/Agree dialogue acts often occur with the Joy emotion, Apology with Sadness, and Thanking with Joy. We make the Emotional Dialogue Acts (EDA) corpus publicly available to the research community for further study and analysis.

Programming robots for a safe interaction with humans is extremely complex especially in collaborative tasks. One reason is the unpredictable behaviour of humans that may have an intention which is not clear to the robot. We present a novel architecture for a safe human-robot collaboration scenario in a shared tabletop workspace based on intuitive multimodal language and gesture instructions and behaviour recognition. In our example scenario, a human and a robot arm collaboratively have to assemble a Tangram puzzle. The configuration space of the robot is constrained by a combination of learned behaviour patterns of the user by tracking its arm and direct audio-visual instructions regarding the sharing of the workspace. This ensures a safe and non-obstructive collaboration behavior of the robot which can constantly be updated during task execution. In this paper, we present initial results with a focus on instruction understanding.

Previous work on emotion recognition demonstrated a synergistic effect of combining several modalities such as auditory, visual, and transcribed text to estimate the affective state of a speaker. Among these, the linguistic modality is crucial for the evaluation of an expressed emotion. However, manually transcribed spoken text cannot be given as input to a system practically. We argue that using ground-truth transcriptions during training and evaluation phases leads to a significant discrepancy in performance compared to real-world conditions, as the spoken text has to be recognized on the fly and can contain speech recognition mistakes. In this paper, we propose a method of integrating an automatic speech recognition (ASR) output with a character-level recurrent neural network for sentiment recognition. In addition, we conduct several experiments investigating sentiment recognition for human-robot interaction in a noise-realistic scenario which is challenging for the ASR systems. We quantify the improvement compared to using only the acoustic modality in sentiment recognition. We demonstrate the effectiveness of this approach on the Multimodal Corpus of Sentiment Intensity (MOSI) by achieving 73,6% accuracy in a binary sentiment classification task, exceeding previously reported results that use only acoustic input. In addition, we set a new state-of-the-art performance on the MOSI dataset (80.4% accuracy, 2% absolute improvement).

Acoustically expressed emotions can make communication with a robot more efficient. Detecting emotions like anger could provide a clue for the robot indicating unsafe/undesired situations. Recently, several deep neural network-based models have been proposed which establish new state-of-the-art results in affective state evaluation. These models typically start processing at the end of each utterance, which not only requires a mechanism to detect the end of an utterance but also makes it difficult to use them in a real-time communication scenario, e.g. human-robot interaction. We propose the EmoRL model that triggers an emotion classification as soon as it gains enough confidence while listening to a person speaking. As a result, we minimize the need for segmenting the audio signal for classification and achieve lower latency as the audio signal is processed incrementally. The method is competitive with the accuracy of a strong baseline model, while allowing much earlier prediction.