Search in the TU Delft Repository Collections
Recently Added Records
Correction to
Computing parameter-dependent invariant manifolds for data-free nonlinear model reduction (Nonlinear Dynamics, (2026), 114, 15, (1012), 10.1007/s11071-026-12814-z)
Correction to: Nonlinear Dynamics (2026) 114:1012 https://doi.org/10.1007/s11071-026-12814-z. The article was originally published without accurate corresponding author information. The publisher acknowledges the errors and asks readers to note the updated article with correct authorship details. The original article has been corrected.
...
Correction to: Nonlinear Dynamics (2026) 114:1012 https://doi.org/10.1007/s11071-026-12814-z. The article was originally published without accurate corresponding author information. The publisher acknowledges the errors and asks readers to note the updated article with correct authorship details. The original article has been corrected.
This study investigates how coordinated rotor tilt and elevon deflection shape the aerodynamic performance and maneuvering capability of a tilt-rotor tailsitter UAV, with particular emphasis on high-speed sharp turns where maximizing lift is essential for minimizing turning radius. Extensive wind tunnel experiments were conducted across a range of airspeeds, angles of attack, throttle settings, and actuator combinations to quantify axial force, lift, and pitching moment while capturing the effects of wing–propeller interaction. Pitching-moment trim tests were also performed by dynamically adjusting actuator input combinations to maintain pitch moment equilibrium. The results show that upward rotor tilt, when combined with downward elevon deflection, produces substantially higher lift while preserving pitch moment equilibrium within the tested positive-AoA forward flight envelope, outperforming either actuation used alone. By embedding the derived empirical models into an equilibrium-constrained sharp turn maneuver, the study further predicts a theoretical minimum turning radius of 8.01 m at 18 m/s under an extrapolated throttle command condition, corresponding to a 31.8% reduction compared to an equivalent fixed-rotor tailsitter with elevons, suggesting that synergistic thrust vectoring and control-surface deflection enable aggressive maneuvers without compromising pitch stability or forward speed.
...
This study investigates how coordinated rotor tilt and elevon deflection shape the aerodynamic performance and maneuvering capability of a tilt-rotor tailsitter UAV, with particular emphasis on high-speed sharp turns where maximizing lift is essential for minimizing turning radius. Extensive wind tunnel experiments were conducted across a range of airspeeds, angles of attack, throttle settings, and actuator combinations to quantify axial force, lift, and pitching moment while capturing the effects of wing–propeller interaction. Pitching-moment trim tests were also performed by dynamically adjusting actuator input combinations to maintain pitch moment equilibrium. The results show that upward rotor tilt, when combined with downward elevon deflection, produces substantially higher lift while preserving pitch moment equilibrium within the tested positive-AoA forward flight envelope, outperforming either actuation used alone. By embedding the derived empirical models into an equilibrium-constrained sharp turn maneuver, the study further predicts a theoretical minimum turning radius of 8.01 m at 18 m/s under an extrapolated throttle command condition, corresponding to a 31.8% reduction compared to an equivalent fixed-rotor tailsitter with elevons, suggesting that synergistic thrust vectoring and control-surface deflection enable aggressive maneuvers without compromising pitch stability or forward speed.
Advancements in computing power and robot control algorithms are making robots increasingly capable. They have become so proficient that companies are now starting to deploy robots on public roads for tasks traditionally performed by humans, such as delivery and inspection. However, like humans, robots are not infallible and will occasionally need assistance if they get stuck or lost. While a human in this situation might ask someone nearby for help, this does not yet seem to be a common strategy for robots, despite its potential to significantly simplify their incident recovery procedures.
This thesis investigates how autonomous robots can communicate their need for help, and how to facilitate the process of helping a robot. It begins with an exploratory design process to evaluate various communication and assistance strategies. To efficiently iterate on these help-seeking methods, a small-scale robotic system was designed and built using off-the-shelf components and 3D-printed parts. This platform was then used to gather insights into how users experience these communication methods in a laboratory environment. Finally, the findings from these tests informed the final design of a larger-scale robot communicating its need for assistance in a public setting. Using this robot, it was found that 12% of random bystanders were willing to and able to assist the robot when stuck in a public space. ...
This thesis investigates how autonomous robots can communicate their need for help, and how to facilitate the process of helping a robot. It begins with an exploratory design process to evaluate various communication and assistance strategies. To efficiently iterate on these help-seeking methods, a small-scale robotic system was designed and built using off-the-shelf components and 3D-printed parts. This platform was then used to gather insights into how users experience these communication methods in a laboratory environment. Finally, the findings from these tests informed the final design of a larger-scale robot communicating its need for assistance in a public setting. Using this robot, it was found that 12% of random bystanders were willing to and able to assist the robot when stuck in a public space. ...
Advancements in computing power and robot control algorithms are making robots increasingly capable. They have become so proficient that companies are now starting to deploy robots on public roads for tasks traditionally performed by humans, such as delivery and inspection. However, like humans, robots are not infallible and will occasionally need assistance if they get stuck or lost. While a human in this situation might ask someone nearby for help, this does not yet seem to be a common strategy for robots, despite its potential to significantly simplify their incident recovery procedures.
This thesis investigates how autonomous robots can communicate their need for help, and how to facilitate the process of helping a robot. It begins with an exploratory design process to evaluate various communication and assistance strategies. To efficiently iterate on these help-seeking methods, a small-scale robotic system was designed and built using off-the-shelf components and 3D-printed parts. This platform was then used to gather insights into how users experience these communication methods in a laboratory environment. Finally, the findings from these tests informed the final design of a larger-scale robot communicating its need for assistance in a public setting. Using this robot, it was found that 12% of random bystanders were willing to and able to assist the robot when stuck in a public space.
This thesis investigates how autonomous robots can communicate their need for help, and how to facilitate the process of helping a robot. It begins with an exploratory design process to evaluate various communication and assistance strategies. To efficiently iterate on these help-seeking methods, a small-scale robotic system was designed and built using off-the-shelf components and 3D-printed parts. This platform was then used to gather insights into how users experience these communication methods in a laboratory environment. Finally, the findings from these tests informed the final design of a larger-scale robot communicating its need for assistance in a public setting. Using this robot, it was found that 12% of random bystanders were willing to and able to assist the robot when stuck in a public space.
Dysarthric speech remains challenging for state-of-the-art automatic speech recognition (ASR) systems, despite their high accuracy on typical speech. Personalized dysarthric ASR offers a way to address the substantial differences in speech characteristics between individuals, but it is constrained by the limited amount of speaker-specific dysarthric data available for training. Speech produced by the same target speaker in another language provides an additional source of labeled data while preserving speaker-specific and dysarthric speech characteristics.
This study investigates bilingual phoneme-level contrastive learning (CL) for personalized dysarthric ASR in a case study involving a single speaker with severe dysarthria producing speech in Dutch and English. Phoneme-level modeling is used as phonemes provide a natural unit for identifying relationships between speech sounds across languages. English and Dutch speech are jointly modeled by constructing contrastive pairs from phonemes that are either shared across the two languages or manually identified as phonetically equivalent. Because these explicitly cross-lingual relationships constitute only a subset of the possible positive pairs, three positive sampling strategies and corresponding loss-weighting variants are evaluated to investigate whether giving them greater influence during training improves recognition or more strongly shapes the learned representation space. Two approaches for incorporating additional typical speech are also evaluated to investigate the effect of training-data composition.
Evaluation considers both phoneme error rate (PER) and the structure of the learned phoneme embedding space using cosine-distance, nearest-neighbor, and silhouette-based analyses. To the best of our knowledge, this is the first study to investigate bilingual phoneme-level contrastive learning for personalized dysarthric ASR.
Results show that phoneme-level CL improves recognition over bilingual fine-tuning using only Connectionist Temporal Classification (CTC), but explicitly prioritizing cross-lingual phoneme relationships does not provide a statistically significant improvement over the simpler same-symbol contrastive learning. At the representation level, bilingual and weighted objectives can bring cross-lingual phoneme representations closer together and produce better-separated phoneme clusters, but these changes do not consistently result in lower PER. Replacing dysarthric speech with typical speech significantly worsens recognition when the amount of training data is kept approximately constant, whereas adding typical speech on top of the full dysarthric training set significantly improves recognition; however, the latter condition also contains more training data, so the improvement cannot be attributed specifically to typical speech. Overall, the results show a benefit from phoneme-level contrastive learning for personalized dysarthric ASR, but no clear additional benefit from explicitly encoding canonical Dutch--English phoneme relationships. This suggests that the usefulness of bilingual contrastive learning may depend on defining cross-lingual relationships that better reflect the target speaker's actual speech patterns. ...
This study investigates bilingual phoneme-level contrastive learning (CL) for personalized dysarthric ASR in a case study involving a single speaker with severe dysarthria producing speech in Dutch and English. Phoneme-level modeling is used as phonemes provide a natural unit for identifying relationships between speech sounds across languages. English and Dutch speech are jointly modeled by constructing contrastive pairs from phonemes that are either shared across the two languages or manually identified as phonetically equivalent. Because these explicitly cross-lingual relationships constitute only a subset of the possible positive pairs, three positive sampling strategies and corresponding loss-weighting variants are evaluated to investigate whether giving them greater influence during training improves recognition or more strongly shapes the learned representation space. Two approaches for incorporating additional typical speech are also evaluated to investigate the effect of training-data composition.
Evaluation considers both phoneme error rate (PER) and the structure of the learned phoneme embedding space using cosine-distance, nearest-neighbor, and silhouette-based analyses. To the best of our knowledge, this is the first study to investigate bilingual phoneme-level contrastive learning for personalized dysarthric ASR.
Results show that phoneme-level CL improves recognition over bilingual fine-tuning using only Connectionist Temporal Classification (CTC), but explicitly prioritizing cross-lingual phoneme relationships does not provide a statistically significant improvement over the simpler same-symbol contrastive learning. At the representation level, bilingual and weighted objectives can bring cross-lingual phoneme representations closer together and produce better-separated phoneme clusters, but these changes do not consistently result in lower PER. Replacing dysarthric speech with typical speech significantly worsens recognition when the amount of training data is kept approximately constant, whereas adding typical speech on top of the full dysarthric training set significantly improves recognition; however, the latter condition also contains more training data, so the improvement cannot be attributed specifically to typical speech. Overall, the results show a benefit from phoneme-level contrastive learning for personalized dysarthric ASR, but no clear additional benefit from explicitly encoding canonical Dutch--English phoneme relationships. This suggests that the usefulness of bilingual contrastive learning may depend on defining cross-lingual relationships that better reflect the target speaker's actual speech patterns. ...
Dysarthric speech remains challenging for state-of-the-art automatic speech recognition (ASR) systems, despite their high accuracy on typical speech. Personalized dysarthric ASR offers a way to address the substantial differences in speech characteristics between individuals, but it is constrained by the limited amount of speaker-specific dysarthric data available for training. Speech produced by the same target speaker in another language provides an additional source of labeled data while preserving speaker-specific and dysarthric speech characteristics.
This study investigates bilingual phoneme-level contrastive learning (CL) for personalized dysarthric ASR in a case study involving a single speaker with severe dysarthria producing speech in Dutch and English. Phoneme-level modeling is used as phonemes provide a natural unit for identifying relationships between speech sounds across languages. English and Dutch speech are jointly modeled by constructing contrastive pairs from phonemes that are either shared across the two languages or manually identified as phonetically equivalent. Because these explicitly cross-lingual relationships constitute only a subset of the possible positive pairs, three positive sampling strategies and corresponding loss-weighting variants are evaluated to investigate whether giving them greater influence during training improves recognition or more strongly shapes the learned representation space. Two approaches for incorporating additional typical speech are also evaluated to investigate the effect of training-data composition.
Evaluation considers both phoneme error rate (PER) and the structure of the learned phoneme embedding space using cosine-distance, nearest-neighbor, and silhouette-based analyses. To the best of our knowledge, this is the first study to investigate bilingual phoneme-level contrastive learning for personalized dysarthric ASR.
Results show that phoneme-level CL improves recognition over bilingual fine-tuning using only Connectionist Temporal Classification (CTC), but explicitly prioritizing cross-lingual phoneme relationships does not provide a statistically significant improvement over the simpler same-symbol contrastive learning. At the representation level, bilingual and weighted objectives can bring cross-lingual phoneme representations closer together and produce better-separated phoneme clusters, but these changes do not consistently result in lower PER. Replacing dysarthric speech with typical speech significantly worsens recognition when the amount of training data is kept approximately constant, whereas adding typical speech on top of the full dysarthric training set significantly improves recognition; however, the latter condition also contains more training data, so the improvement cannot be attributed specifically to typical speech. Overall, the results show a benefit from phoneme-level contrastive learning for personalized dysarthric ASR, but no clear additional benefit from explicitly encoding canonical Dutch--English phoneme relationships. This suggests that the usefulness of bilingual contrastive learning may depend on defining cross-lingual relationships that better reflect the target speaker's actual speech patterns.
This study investigates bilingual phoneme-level contrastive learning (CL) for personalized dysarthric ASR in a case study involving a single speaker with severe dysarthria producing speech in Dutch and English. Phoneme-level modeling is used as phonemes provide a natural unit for identifying relationships between speech sounds across languages. English and Dutch speech are jointly modeled by constructing contrastive pairs from phonemes that are either shared across the two languages or manually identified as phonetically equivalent. Because these explicitly cross-lingual relationships constitute only a subset of the possible positive pairs, three positive sampling strategies and corresponding loss-weighting variants are evaluated to investigate whether giving them greater influence during training improves recognition or more strongly shapes the learned representation space. Two approaches for incorporating additional typical speech are also evaluated to investigate the effect of training-data composition.
Evaluation considers both phoneme error rate (PER) and the structure of the learned phoneme embedding space using cosine-distance, nearest-neighbor, and silhouette-based analyses. To the best of our knowledge, this is the first study to investigate bilingual phoneme-level contrastive learning for personalized dysarthric ASR.
Results show that phoneme-level CL improves recognition over bilingual fine-tuning using only Connectionist Temporal Classification (CTC), but explicitly prioritizing cross-lingual phoneme relationships does not provide a statistically significant improvement over the simpler same-symbol contrastive learning. At the representation level, bilingual and weighted objectives can bring cross-lingual phoneme representations closer together and produce better-separated phoneme clusters, but these changes do not consistently result in lower PER. Replacing dysarthric speech with typical speech significantly worsens recognition when the amount of training data is kept approximately constant, whereas adding typical speech on top of the full dysarthric training set significantly improves recognition; however, the latter condition also contains more training data, so the improvement cannot be attributed specifically to typical speech. Overall, the results show a benefit from phoneme-level contrastive learning for personalized dysarthric ASR, but no clear additional benefit from explicitly encoding canonical Dutch--English phoneme relationships. This suggests that the usefulness of bilingual contrastive learning may depend on defining cross-lingual relationships that better reflect the target speaker's actual speech patterns.
Multi-agent systems increasingly rely on communication and orchestration to coordinate, yet their interactions are often transient rather than accumulated into persistent relationships. We ask whether persistent relational structure can instead emerge from interaction itself, and formalise a neuron–agent isomorphism that models a MAS as a social graph evolving through reward-modulated Hebbian plasticity. The graph is coupled back into agent behaviour through reward diffusion and weight-gated experience sharing for RL-fine-tuned agents, and through an inference-time social module for VLM-based agents. We evaluate the framework in WIRE, a five-stage embodied Craftium environment with increasing coordination demands. We characterise when social plasticity supports coordination, showing that its effect depends on the underlying agent and the choice of co-firing signal. On Gemma-E4B, Hebbian coupling increases mean task return from 521 to 570 and cooperative milestone completion from 8.5% to 12.1%, with longer-timescale reward modulation further increasing cooperative completion to 14.4%. Decentralised plasticity-based coordination is also competitive with centralised orchestration at matched compute. Different social cues induce distinct relational structures, with observation yielding the highest cooperative completion (16.3%) despite communication producing substantially stronger bonds. Finally, when initial bond strength is controlled, agents preferentially collaborate with teammates with whom they had previously co-fired (0.56 vs. 0.40 under random re-pairing) and retain stronger bonds, showing that agents that co-fire together wire together even after team recomposition. The code can be found at: https://github.com/tapri-lab/wired-together.
...
Multi-agent systems increasingly rely on communication and orchestration to coordinate, yet their interactions are often transient rather than accumulated into persistent relationships. We ask whether persistent relational structure can instead emerge from interaction itself, and formalise a neuron–agent isomorphism that models a MAS as a social graph evolving through reward-modulated Hebbian plasticity. The graph is coupled back into agent behaviour through reward diffusion and weight-gated experience sharing for RL-fine-tuned agents, and through an inference-time social module for VLM-based agents. We evaluate the framework in WIRE, a five-stage embodied Craftium environment with increasing coordination demands. We characterise when social plasticity supports coordination, showing that its effect depends on the underlying agent and the choice of co-firing signal. On Gemma-E4B, Hebbian coupling increases mean task return from 521 to 570 and cooperative milestone completion from 8.5% to 12.1%, with longer-timescale reward modulation further increasing cooperative completion to 14.4%. Decentralised plasticity-based coordination is also competitive with centralised orchestration at matched compute. Different social cues induce distinct relational structures, with observation yielding the highest cooperative completion (16.3%) despite communication producing substantially stronger bonds. Finally, when initial bond strength is controlled, agents preferentially collaborate with teammates with whom they had previously co-fired (0.56 vs. 0.40 under random re-pairing) and retain stronger bonds, showing that agents that co-fire together wire together even after team recomposition. The code can be found at: https://github.com/tapri-lab/wired-together.