Tracking
-
Onboard Vision and MPC Navigation for Underwater Robots: An Open BlueROV2 Platform for Multi-Robot Experiments & Docking
2026-09-29 20:33:23Victor Nan Fernandez-Ayala, Wiktor Kowalczyk, Cezary Banaszek, Dimos V. DimarogonasAbstract
Autonomous underwater robots require robust perception, estimation and control to operate in confined environments. This paper presents an open-source BlueROV2 platform combining onboard vision with nonlinear Model Predictive Control (NMPC) for autonomous navigation and docking. The platform integrates an NVIDIA Jetson Orin NX and an Intel RealSense D435i stereo camera in a modular pressure housing. Underwater-calibrated stereo depth and realtime object detection provide relative position measurements of nearby BlueROV2 vehicles in the camera and body frames. A quaternion-based estimator fuses external pose and inertial measurements, while an NMPC controller based on a nonlinear six-degree-of-freedom model tracks planned navigation and docking trajectories. To support reproducible development, we also provide open-source physics-based PX4 SITL and Gazebo environments, multi-robot simulation tools and a lowcost physical docking station. Experiments evaluate underwater perception, onboard computational performance, state estimation, trajectory tracking and autonomous docking.
Abstract (translated)
URL
https://arxiv.org/abs/2609.38511
PDF
-
L1-MPPI: L1 Adaptive Model Predictive Path Integral for Agile UAV Control
2026-09-29 19:59:05Luk\'a\v{s} Kotek, Ond\v{r}ej Proch\'azka, Vojt\v{e}ch Von\'asek, Martin Saska, Robert P\v{e}ni\v{c}kaAbstract
This work proposes the L1 Adaptive Model Predictive Path Integral (L1-MPPI). It cascades L1 adaptive control with the Model Predictive Path Integral (MPPI) to improve tracking of high-speed UAV trajectories. Thanks to the L1augmentation, the tracking remains accurate even under model uncertainties and external disturbances, such as an additional payload or a mismatch in the modeled aerodynamic drag. In contrast to existing MPPI approaches for UAV control that do not explicitly model aerodynamic effects, varying payloads, and typically neglect the dynamics of low-level motor controllers, our L1-MPPI approach enhances the dynamic model used in the MPPI by incorporating the low-level flight controller and motor dynamics, as well as an iterative mixing scheme that reflects the approach of the low-level controller. The proposed method demonstrates improved tracking performance in both simulation and the real world, even when the UAV is subjected to an unknown payload. In flight with 35% mass increase, our approach lowers the RMSE by 58.61% with respect to plain MPPI. Compared to the same MPPI using an online mass estimator in place of the L1 augmentation, the RMSE is lower by 38.59%. During the real-world experiments the UAV reaches speeds up to 13.50 m/s and accelerations up to 2.5 g.
Abstract (translated)
URL
https://arxiv.org/abs/2609.38467
PDF
-
TrackFish3D: Self-Supervised 3D Tracking of Schooling Fish from Multi-view Videos
2026-09-29 18:10:09Patt Phurtivilai, Zhiyang Dou, Yifan Wu, Kinfung Chu, Yuan Liu, Lei Yang, Wenping Wang, Taku KomuraAbstract
Quantifying collective fish behavior requires accurate trajectories, yet multi-view 3D tracking remains challenging due to frequent occlusions, visually similar individuals, and the long-standing scarcity of identity annotations. We present TrackFish3D, a geometry-driven self-supervised framework for dense multi-camera 3D tracking of schooling fish. Instead of relying on appearance-based re-identification or manually annotated identities, TrackFish3D turns calibrated multi-view geometry into supervision: triangulation and reprojection consistency provide pseudo-associations, while a geometric encoder and global association transformer learn all-to-all cross-view correspondence within each frame. To make these associations identity-aware, TrackFish3D introduces a self-supervised contrastive objective that separates co-visible individuals in the embedding space, together with a temporal predictor that preserves identities and bridges short occlusions across frames. The resulting model is trained once on unlabeled footage and applied directly to unseen test videos, requiring no cross-view identity labels, temporal annotations, 3D ground truth, appearance features, or test-time optimization. On our benchmark, TrackFish3D improves 3D Multi-Object Tracking Accuracy from 87.7% for the strongest baseline to 95.8%. On the 3D-ZeF zebrafish benchmark, it achieves 81.1% MOTA, compared with 77.4% for the best geometric baseline. TrackFish3D also generalizes beyond fish, achieving strong results on real-world bird tracking.
Abstract (translated)
URL
https://arxiv.org/abs/2609.38347
PDF
-
Pretraining Latent Information Feedback Transformers with Teacher Supervision
2026-09-29 17:57:40Dor Tirosh, Ido Amos, Mor GevaAbstract
Transformer language models (LMs) are feed-forward: deep-layer representations are never fed back to shallower layers, and the only pathway for information to flow downward across generation steps is the decoded token. This narrow channel forces models to recompute intermediate results and to discard alternative continuations. In this work, we remove this bottleneck during pretraining, introducing the LIFT (Latent Information Feedback Transformer) architecture and training method which enable LMs to propagate state across generation. We achieve this by turning recurrent-state learning into a teacher-forced prediction problem: each input token is paired with an information-dense state, derived from the next-token distribution of an off-the-shelf pretrained LM. The model, extended with a small number of additional parameters, is then trained to predict both the next token and the next state. As the input states are precomputed, pretraining remains fully parallel across positions. At inference, the model's own predicted states are fed back, with a minor computational overhead that decreases with model size. Experiments with pretrained models ranging from 135M to 1B parameters show that LIFT consistently outperforms standard Transformers and baselines on language modeling, downstream reasoning tasks, and procedural tasks under token-matched budget, while being on par with or ahead of compute-matched Transformers. Moreover, a controlled study on a state-tracking task shows that a tiny LIFT outperforms same-size Transformers trained on 8x more data, even when trained with the states of a Transformer that fails the task. Overall, we show that LMs can learn to exploit deep-to-shallow feedback during pretraining via scalable teacher supervision.
Abstract (translated)
URL
https://arxiv.org/abs/2609.38149
PDF
-
CrossBFM: Distilling a Shared Latent Behavior Space Across Humanoid Embodiments
2026-09-29 17:40:11Tan-Dzung Do, Tuan Dat Phuong, Nico Bohlinger, Cuc T. Trinh, Siwei Ju, Vien Anh Ngo, Jan Peters, Xinchao Wang, An T. LeAbstract
Behavior Foundation Models (BFMs) give humanoids a promptable policy over a latent behavior space, enabling one single vector to represent a motion to imitate, a pose to reach, or a reward to maximize. Forward-Backward representations successfully produce such spaces, but at the cost of hundreds of GPU-hours for a single robot. Moreover, when the training process is repeated for a second robot, it produces a second space unrelated to the first, resulting in embodiment-specific latents that do not unify or transfer. We address these problems with CrossBFM, treating the latent space as the transferable asset for various embodiments. As retargeting provides frame-level cross-embodiment correspondence, we propose a unified encoder architecture with no robot-specific parameters for distilling the behavior space to address all training embodiments simultaneously in less than a GPU-hour. Following this encoder, latent-conditioned trackers turn the distilled latent into whole-body control in a conventional PPO training manner in just 10 more GPU-hours. On three distilled humanoids, all three prompting modes transfer: motion tracking with latent-conditioned policy losing only $0.025$ rad to its joint-conditioned counterpart, smooth goal reaching between poses with no falls, and reward optimization for all $41$ reward prompts. Our experiments further reveal that 1) regressing the encoder on a quarter of the motion corpus costs only $5\%$ of tracking performance and 2) training the encoder on a subset of robots and evaluating on an unseen one recovers up to $89\%$ of the tracking performance of seen robots, demonstrating cross-embodiment generalization to morphologically similar robots. We also verify the pipeline on real robots across all three prompting modes and with flow-based generated latents. Project website: this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2609.38087
PDF
-
Pow3R-SLAM: Real-Time RGB-D SLAM with 3D Reconstruction Priors
2026-09-29 17:22:24Christopher Kolios, Ishaan Mehta, Sasa Janjic, Yeganeh Bahoo, Sajad SaeediAbstract
We present Pow3R-SLAM, a real-time RGB-D simultaneous localization and mapping (SLAM) system that uses Pow3R for tracking and mapping. Inspired by MASt3R-SLAM, a recent work on monocular SLAM using two-view 3D reconstruction priors, we extend the work to incorporate depth as a prior on the network's prediction, rather than as geometry to fuse. Where traditional RGB-D SLAM systems struggle with sparsity in the depth images, Pow3R utilizes the available depth to give a better-conditioned pointmap, while inferring the depths in empty regions from the two-view photometric, depth, and intrinsic data. Evaluated against MASt3R-SLAM following its protocol on 24 sequences from TUM, 7-Scenes, and Replica, Pow3R-SLAM runs 1.6x faster in wall time, has 15% lower mean trajectory error, a 3.1x lower unscaled error, and produces denser maps, with a 30% lower Chamfer distance. We also introduce a hybrid variant that runs 2.1x faster than MASt3R-SLAM at 25.3 frames per second (FPS), while maintaining improved tracking and mapping accuracy. Against ORB-SLAM3 in RGB-D mode, Pow3R-SLAM is more accurate on TUM, 7-Scenes, and ETH3D-SLAM, and completes every TUM sequence. While Pow3R-SLAM can struggle on a small set of self-similar scenes, its overall performance shows that adding depth as a prior for two-view 3D reconstruction SLAM can be beneficial. A project webpage is available at: this https URL , and code will be made open-source upon acceptance.
Abstract (translated)
URL
https://arxiv.org/abs/2609.38054
PDF
-
Does Local Video Understanding Transfer Across Encounters? The EgoGears Benchmark
2026-09-29 16:20:01Yuedong Tan, Lei Qi, Yu Liu, Di Wen, Ruiping Liu, Xiaoye Wang, Yufan Chen, Junwei Zheng, Chengzhi Wu, Chen Zhang, Zhihang Chen, Haiwen Sun, Zongwei Wu, Radu Timofte, Danda Pani Paudel, Kunyu PengAbstract
Embodied systems must make knowledge acquired during one encounter usable in another despite changes in viewpoint, motion, and illumination. Yet aggregate cross-video accuracy conflates failures of local perception with failures to preserve observation identity, establish correspondence, and compose evidence, obscuring whether local video understanding actually transfers. We introduce EgoGears, a complementary single- and multi-video benchmark designed to diagnose this transition. It contains 567 single-video and 1,487 multi-video questions derived from 126 human-collected egocentric recordings covering 39 outdoor routes. Repeated traversals across movement speeds and lighting conditions ground comparisons in shared physical environments; 531 questions require alignment across independent recordings. Single-video questions measure the local visual, spatial, and motion evidence available to a model, while multi-video questions test whether evidence remains bound to the correct observation and can be composed into consistent route relationships. We report 29 single-video and 31 multi-video MLLM configurations across six model families in the main leaderboard. Among the 20 configurations evaluated comparably on both splits, every model performs worse on multi-video questions, with a mean decrease of 22.5 percentage points, and the gap persists when answer format and scoring are held fixed. The gap is not explained simply by additional videos or recording boundaries. The central bottlenecks are observation--evidence binding and ordered route-state tracking. The code and benchmark are publicly available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37938
PDF
-
ByteTraX: Enhancing the ByteTrack Architecture with Optimised Thresholding
2026-09-29 15:23:06Thomas A. O'Shea-WhellerAbstract
The ByteTrack algorithm is a widely used and computationally efficient multi-object tracking architecture. Its core innovation lies in the combination of lenient bounding box associations with tracklet similarity matching to robustly deal with object occlusions. However, this strategy is nevertheless vulnerable to erroneous track reclassification and identity switching, as detection confidence scores dictate association priority. To address this, I present a simple enhancement of the ByteTrack architecture, named ByteTraX, that optimises track continuity via a single unified matching threshold, while penalising identity switches through stringent track initiation criteria. This approach achieves consistently improved performance across a range of diverse benchmarks including GMOT-40, LC-MOT, SportsMOT, TeamTrack, DAMUNT, and DeepSea-MOT, while simultaneously increasing processing speed by >10%. Specifically, results demonstrate a >40% reduction in identity switches, accompanied by mean increases in HOTA of 3.6, IDF1 of 5.6, and FPS of 6.3. As such, adoption of the ByteTraX algorithm has the potential to substantially enhance tracking performance over the ByteTrack baseline, while retaining the efficiency needed for real-time deployment. To facilitate usage, I provide the source code, integration functionality for the YOLO family of object detection models, and deployment instructions via an open source repository.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37801
PDF
-
Generative Interactions: Weaving Multiparty Human Motion with Bilevel Latent Dynamics
2026-09-29 14:40:40Ojas Shirekar, Yash Surange, Agustinas Ju\v{c}as, Chirag RamanAbstract
Human social behaviour is not a collection of independent motions, but a jointly organised process in which group dynamics and individual variation continuously shape one another. Yet existing social motion models often prioritise plausible trajectories while leaving interaction state implicit, limiting their ability to transfer across groups, tasks, and partial-observation regimes. To address this gap, we introduce Bilevel Representations for Agent Interaction Dynamics (BRAID), a hierarchical sequential latent-variable model for generative multi-person interaction. BRAID explicitly formulates social motion generation as a meta-transfer learning problem: shared interaction priors are learned across datasets and adapted through arbitrary context sets of observed people and joints. The model represents each scene through a group-level latent state that captures shared interaction dynamics and person-level latent states that capture individual behaviour conditioned on the evolving group context. This modelling choice enables coherent generation under full, sparse, or partial observations while exposing compact social-state vectors that can serve as an interface for downstream embodied-agent systems. We evaluate BRAID under a unified SMPL-based representation on social forecasting, tracking and in-filling, and response generation, using metrics that assess not only reconstruction accuracy but also realism, diversity, temporal alignment, and interpersonal coordination. We further analyse the hierarchical latent space, showing that it captures separable group- and individual-level structure.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37708
PDF
-
Reader Proficiency Shapes Layer-wise Surprisal Profiles
2026-09-29 14:32:50Akio Hayakawa, Horacio SaggionAbstract
Reading behaviour varies not only with linguistic input, but also with reader proficiency. In this study, we investigate whether the layer-wise relationship between surprisal from large language models (LLMs) and human gaze behaviour differs across readers with different levels of proficiency and across gaze measures. Using eye-tracking data from the MECO L2 corpus, we compare readers with high and low vocabulary proficiency on first-pass gaze duration (FPGD) and total gaze duration (TGD). We quantify the distribution of the predictive power of surprisal across model layers using Predictive Depth. Across 12 tested LLMs, we find that readers with lower vocabulary proficiency tend to show deeper Predictive Depth for FPGD, while this difference is smaller for TGD. Also, TGD itself shows deeper Predictive Depth than FPGD in both proficiency groups. These patterns suggest that where predictive power is concentrated across LLM layers may be related to the timing and breadth of the reading processes captured by different gaze measures, and that this relationship can vary with reader proficiency. Our leave-one-out analysis further shows that the advantage of informative internal layers extends to unseen texts, although the practical improvements in prediction are limited. Overall, our results show that layer-wise LLM surprisal provides a useful perspective on variation in reading behaviour across both reader groups and gaze measures.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37688
PDF
-
Learning Expressive and Compositional Motion Representation via Spectral Skills
2026-09-29 14:27:49Feiyang Wu, Chenxiao Gao, Chen Yang, Ye Zhao, Bo Dai, Anqi WuAbstract
Robotic foundation models offer a promising path toward general-purpose humanoid robot control, often through hierarchical architectures. However, their effectiveness depends on the command interface between the planner and the controller, which must support accurate execution while remaining easy to predict, and ideally allow new behaviors to be composed from prior ones. In this work, we introduce spectral skills, a latent representation of this interface that meets these requirements through predictive representation learning. By design, spectral skills compactly encode short motion segments and are learned by predicting subsequent motion rather than reconstructing the encoder input. On a 29-DoF humanoid, a controller conditioned on spectral skills reduces global tracking error by 62\% relative to the state of the art. The same frozen controller chains independently encoded skills without a separate transition policy. It also composes new behaviors by adding orthogonal directions to any compatible base skill, producing combinations unseen in the training data. We demonstrate tracking, chaining, and composition, as well as control through a language-conditioned planner, on Unitree G1 hardware. Project page: this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2609.37677
PDF
-
LLMs learn different forms of metacognition when trained to predict their own accuracy
2026-09-29 13:00:25Nicolas Yax, Stefano Palminteri, Pierre-Yves OudeyerAbstract
Large language models are trained to always produce an answer, regardless of whether they possess the relevant knowledge, which leads them to fabricate facts. Prior work has shown that LLMs' confidence estimates correspond poorly to their actual performance, and that fine-tuning can substantially improve them. However, what models actually learn during such training remains poorly understood. We investigate how LLMs acquire metacognitive monitoring, the ability to know what one knows, by training 10 open-weight LLMs to predict their own accuracy on factual multiple-choice questions before answering them. We find that trained confidence reflects two distinct signals. While on questions close to the training data, it tracks the model's true accuracy, in other domains, it instead tracks output consistency: the concentration of the model's answer distribution. Output consistency tracking emerges early in training and generalizes across datasets, whereas accuracy tracking develops later and remains local to the training distribution. These results suggest that calibration training may not teach models to generally detect errors they commit confidently, and they raise broader questions about the nature of metacognition in artificial systems.
Abstract (translated)
URL
https://arxiv.org/abs/2609.33886
PDF
-
UGO: Unified Architecture for General Multi-Object Tracking by Segmentation
2026-09-29 12:09:56Jer Pelhan, Alan Lukezic, Matej KristanAbstract
General multi-object tracking (GMOT) tracks all instances of a user-specified category from a single first-frame exemplar. Prior work relies on bounding boxes and surrogate training, and struggles with non-rigid objects, crowded scenes, and distractors. We introduce UGO, a unified GMOT tracker that pairs a pretrained exemplar-conditioned detection head with an instance-propagation head in a common architecture. A novel training-free, energy-minimization consolidation method converts overlapping proposals into exclusive pixel-wise masks and detections, resolving over-segmentation, duplicates, and conflicts. A hierarchical memory spanning global and instance levels improves recall and per-instance segmentation accuracy using a new memory management protocol. UGO sets a new state-of-the-art on GMOT benchmarks and video object counting, and is competitive with specialist MOT methods, establishing a strong paradigm for unified, open-category multi-object tracking.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37339
PDF
-
Battery-Aware Reinforcement Learning for Aggressive Quadrotor Flight
2026-09-29 11:52:34Alejandro Sanchez Roncero, Olov Andersson, Petter OgrenAbstract
Agile flight tasks such as drone racing and pursuit-evasion require strong acceleration and precise turns, but the available thrust changes as the battery discharges and voltage drops under load. Conservative command limits make this variation easier to tolerate, at the cost of unused performance. We investigate how learned controllers can use that additional thrust while retaining the flight controller's voltage compensation and rate control. Our training simulator couples an identified load-transient battery model to rotor dynamics and firmware saturation. The feedforward policy receives filtered voltage during both training and deployment. Controlled ablations distinguish the benefit of a larger thrust-command range from that of voltage information. On a 38 g Crazyflie Brushless, the resulting policy reduces circle tracking error by 49% relative to stock-authority RL at 3.84 m/s, while preserving easy-task precision. Mean 20-lap race time decreases from 106.22 s to 95.24 s. Compared with a voltage-blind policy with the same increased authority, hardware error and race time are lower by 15.3% and 4.5%, respectively. In simulation, replacing the policy's voltage input with a recording from a different battery condition worsens hard-circle tracking, with a smaller, voltage-dependent effect in racing. Together, these results show where a simple voltage input complements existing actuator compensation in aggressive learned flight.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37316
PDF
-
CST-WM: A Causally Structured World Model for Embodied Visual Tracking
2026-09-29 10:26:07Junyi Hu, Shuaihang Yuan, Jiazhao Liang, Yi FangAbstract
Embodied visual tracking requires a robot to choose actions that keep a moving target observable at a suitable distance, and to recover it after occlusion, out-of-view drift, or distractor crossings. We cast the task as planning over future target evidence with an action-conditioned world model. In logged tracking data, however, the behavior policy's actions are correlated with where the target is, so a generic predictor can learn a shortcut: it writes the current action directly into its prediction of target evidence, instead of letting the action affect that evidence only by moving the robot and changing what it observes. We call this failure causal hallucination; the resulting rollouts look plausible but rank candidate actions for the wrong reason. We propose CST-WM, a causally structured world model whose state is split into target-evidence, robot, and observation branches. Its transition removes the same-step edge from action to target evidence but keeps the path through robot motion and the resulting views, so candidate actions are still distinguished by their predicted ego-motion. With rollout-based model-predictive control, a single model handles both steady following and re-acquisition after target loss. On EVT-Bench and Habitat 3.0, covering standard tracking, target-loss recovery, and cross-dataset transfer, CST-WM improves following, distance-range control, safety, and re-acquisition over reactive trackers and world-model baselines, and removing the action mask causes the largest drop in re-acquisition among our ablations. Offline, CST-WM has lower multi-step rollout error, and its ranking of candidate actions agrees better with the simulator's. On a Unitree Go2 quadruped, CST-WM succeeds in 20 of 30 real-world trials under occlusion, distractor crossing, and fast motion, against 14 for TrackVLA.
Abstract (translated)
URL
https://arxiv.org/abs/2609.06302
PDF
-
End-to-End Self-Supervised RGB-T Tracking without Modality Misleading
2026-09-29 09:52:41Shenglan Li, Rui Yao, Kunyang Sun, Hong Jia, Yong Zhou, Javen Qinfeng Shi, Xinyu ZhangAbstract
RGB-T object tracking leverages the complementary characteristics of visible and thermal infrared modalities to improve robustness under adverse conditions. Existing supervised methods typically rely on costly modality-aligned bounding box annotations, while most self-supervised approaches follow a two-stage pseudo-labeling paradigm, making tracker training sensitive to pseudo-label quality and preventing joint end-to-end optimization. In this paper, we propose ESMTrack, a fully end-to-end self-supervised RGB-T tracking framework without offline pseudo-label generation or dense frame-level bounding box annotations. Given only the standard initial-frame annotation used in visual tracking, ESMTrack learns discriminative and temporally consistent representations through two complementary objectives: a grounding triplet loss on annotated initial frames and a cross-frame temporal triplet loss on unlabeled search frames, with reliable samples selected by forward-backward consistency. To address modality dominance bias, ESMTrack employs a three-branch architecture consisting of a fusion branch and two unimodal branches for RGB and thermal inputs. We quantify modality contributions using the Average Peak-to-Correlation Energy by measuring response discrepancies between the fusion and unimodal branches. The resulting reliability estimates guide a training-time modality decoupling mechanism that suppresses dominant-modality shortcuts and adaptively weights cross-modal contrastive learning for task-level alignment. Extensive experiments on five RGB-T tracking benchmarks show that ESMTrack achieves competitive state-of-the-art performance, strong cross-dataset generalization, and real-time inference speed. The source code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37162
PDF
-
All You Need Is Low Fidelity: Zero-Shot Sim-to-Real of Learned Robotic Fish Control
2026-09-29 08:23:18Liam Maloney, Simon Ramchandani, Mike Y. Michelis, Ronan Hinchet, Robert K. KatzschmannAbstract
Complex tasks for underwater robots remain limited by the capabilities of their controllers. Learning a better one for a soft, underactuated robotic fish trades simulator cost against fidelity. We show that an intentionally low-fidelity simulator is enough: a stateless, quasi-steady fluid model with no wake and no added-mass history suffices to learn a \emph{general}, closed-loop controller that transfers to hardware without tuning. Our platform is a soft, single-motor, tendon-driven fish whose policy observes only what the hardware can measure. A staged pipeline grounds the simulator in two independent identifications, fixing the tail dynamics and a stateless fluid model; the policy then acts through a band-limited rhythmic trajectory generator rather than commanding the tail directly. Deployed unchanged in an outdoor pool, a single policy performs closed-loop target reaching, disturbance rejection, and out-of-distribution target acquisition and tracking. The transfer rests on the constraint rather than the fidelity: the generator cannot leave the band over which the fluid was identified. This raises the question of how much of the physics can reside in the controller rather than in the simulator.
Abstract (translated)
URL
https://arxiv.org/abs/2609.36993
PDF
-
MultiTalk: Scaling Full-Duplex Speech Models to Long, Multi-Party, Bilingual Conversation
2026-09-29 07:25:29Ke Wang, Houxing Ren, Zimu Lu, Yunqiao Yang, Zhuofan Zong, Mingjie Zhan, Hongsheng LiAbstract
End-to-end full-duplex speech models have brought open-source machine conversation closer to human-like interaction, yet existing systems remain limited in two intertwined dimensions: long-context robustness and multi-party interaction. Real-world scenarios such as meetings, group lessons, and social-robot reception require a single model to track, contextualize, and respond to multiple speakers over extended durations. Progress is constrained by both data and evaluation: open multi-party speech corpora remain small and are not designed for codec-frame-level full-duplex modeling, while existing long-audio benchmarks focus on passive listening and speech-to-speech benchmarks are mostly short and dyadic. We extend the Moshi paradigm jointly along the long-horizon and multi-party axes in English and Chinese. First, we release 57.6k hours of synthetic training data ($\href{this https URL}{MultiTalkPT}$ and $\href{this https URL}{MultiTalkFT}$) for long-form, multi-party, English-Chinese full-duplex dialogue, with controllable length, participant count, turn-taking, overlap, backchannels, interruptions, addressee shifts, and long-range coreference. Second, we introduce $\href{this https URL}{MultiTalkBench}$, built from real human recordings, for evaluating long-form, multi-party, bilingual full-duplex dialogue. Conversations average 32.6 minutes and include probes for long-range entity tracking, topic coherence, and addressee selection. Third, we train a bilingual Moshi-style model that sustains coherent multi-party English-Chinese conversations over extended durations and substantially outperforms open-source baselines including Moshi, MiniCPM-o-4.5, and Qwen3-Omni-30B-A3B-Instruct on MultiTalkBench.
Abstract (translated)
URL
https://arxiv.org/abs/2609.36903
PDF
-
State Trace Rationale As Auxiliary Task in Reinforcement Learning
2026-09-29 07:07:22Muhammad U. Nasir, Alex Vogt, Steven D. James, Julian TogeliusAbstract
We propose STRAT, an auxiliary task that trains deep reinforcement learning (RL) agents to predict a short textual trace of their own state. Inspired by human spatial navigation, the description combines landmark, route, and survey knowledge, tracking the agent's position, inventory, goals, and immediate progress. Environment rules generate this text online without human labelling. Our method adds a single auxiliary head to a standard policy. Across 60 sparse-reward XLand-MiniGrid tasks, STRAT solves complex environments where standard RL fails outright, while compacting state representations and preventing rank collapse. Beyond performance gains, the predicted trace provides a readable account of agent beliefs at every step for no extra cost.
Abstract (translated)
URL
https://arxiv.org/abs/2609.36867
PDF
-
Scale-Invariant Manipulability Shape Tracking Across Heterogeneous Manipulators
2026-09-29 05:56:41Geunwoo Kwon, Dong-gyu Lee, Kai Li, Soonwoong Hwang, Wansoo KimAbstract
When transferring manipulability across systems with different sizes and kinematic structures, matching absolute ellipsoid scale may be unnecessary when the goal is to reproduce orientation and semi-axis length ratios. Full-matrix tracking, however, penalizes both shape and absolute-scale differences, even when only shape matching is required. We therefore propose a scale-invariant manipulability shape-tracking method that treats matrices differing only by a positive scalar factor as equivalent and uses their unit-determinant representatives. We derive the differential of the unit-determinant shape representative and an orthonormal coordinate representation of the tangent tracking residual under the affine-invariant Riemannian metric (AIRM). The resulting scale-invariant objective is integrated with position and end-effector direction tasks in a constrained joint-velocity quadratic program. Simulations with four heterogeneous robots evaluate robot-to-robot and human-to-robot transfer. On three followers, the proposed method achieves endpoint shape distances of 9.30 x 10^-5 without scale tuning. With robot-specific target scales tuned during motion, the Full method retains endpoint axis-ratio errors of 0.19-0.31 on KR500 and UR20. For human reaching with concurrent tasks, the proposed method yields dual force shapes elongated along X like the human target on all four robots, with endpoint position errors of 2.4-5.6% of reference arm length versus up to 75% for the Full method tracking the original human ellipsoid.
Abstract (translated)
URL
https://arxiv.org/abs/2609.36784
PDF
