Learned humanoid controllers are usually trained in one simulator and then checked in a second one before they go near hardware. We ask what that check says about command delay. We train motion-tracking policies for a 24-joint humanoid on five motions taken from monocular video and test them in an independent CPU MuJoCo harness, which we first verify against the training model. Every policy completes its motion in at least 26 of 30 episodes under the training randomization. With 40 ms of constant command delay, however, survival drops to 15% on average (0% - 45%), and the delay margin is only 20 - 30 ms. Randomizing the delay during training (0 - 50 ms) lifts survival at 40 ms to 95% on average (75% - 100%) and moves the margin to 50 - 70 ms. A control arm trained for the same number of extra iterations without delay does not improve, so the gain comes from the randomization itself. This holds for all five motions with three training seeds each, and a policy trained in PhysX behaves the same way. The price is slightly looser tracking (0.02 - 0.06 rad on four motions) and, for the most dynamic motion, less robustness to other perturbations. We also show that the first sim-to-sim gap we saw came from the harness, a missing floor and a wrong timestep, and not from the physics engines. The 0 - 20 ms delay ranges common in locomotion training lie inside the margin an unrandomized tracker already has. The benefit comes from randomizing beyond it, so the range should be set from the robot's measured latency, and delay margin belongs next to other sim-to-sim metrics.
https://arxiv.org/abs/2610.32277
Vision-language-action (VLA) models are typically pretrained on broad, diverse datasets and subsequently fine-tuned on a comparatively small dataset tailored to a specific target robot. Such adaptation recipe is now widely established, achieving its most mature results on fixed-base arms due to their static base dynamics and consistent reference frame. Mounting the same arm on a quadruped extends the workspace to the entire floor, but fundamentally changes the embodiment: rather than controlling the arm in isolation, the policy must jointly coordinate base navigation, body posture, and end-effector motion. In this work, we study the practical demands of this embodiment change across data collection, model adaptation, and inference design to achieve closed-loop loco-manipulation on a Unitree Go2 with an AgileX Piper arm. We report the choices that made this real-robot deployment feasible, namely model variant, action horizon, and demonstration conventions, together with a study tailored to a pick-and-place task, separating data coverage from model capacity. We show that expanding demonstration coverage reduces held-out end-effector error by 30%, and incorporating secondary skills (such as pushing rather than grasping) introduces new behavioral modes without degrading baseline performance. Finally, closed-loop real-world experiments show how dataset composition shapes success and failure, validating the pipeline for more complex loco-manipulation.
https://arxiv.org/abs/2610.32271
Humanoids increasingly demand lightweight leg mechanisms for agile, dynamic, and energy-efficient locomotion. Reducing distal inertia through proximal actuation has thus become a common design goal. Yet, compact transmissions that preserve torque capability and efficient force transmission remain hard to achieve. This paper presents a novel series-parallel leg design combining a knee parallel mechanism, an ankle parallel mechanism, and a nonlinear knee-ankle coupling where all actuators placed closer to the hip. As the knee flexes, the coupling passively assists ankle dorsiflexion for foot clearance and, as it extends it assists plantarflexion for push-off, redistributing ankle load through the knee transmission. We derive the coupling's nonlinear transmission models and validate them in simulation and on hardware. The design shows that the shared transmission lets both joints draw on each other's actuator capacity by easing the knee's peak torque demand under load and the ankle's peak speed demand during fast motion. With the leg's proximal actuator placement, the leg achieves an effective reflected mass of 0.37 kg at the foot for a 6 kg leg, demonstrating very low distal leg inertia for dynamic humanoid robots.
https://arxiv.org/abs/2610.16604
Protection levels (PLs) bound position errors under specified fault hypotheses and risk budgets, but advanced re-ceiver autonomous integrity monitoring (ARAIM) is formulated for snapshot GNSS estimation rather than the correlated poste-rior of tightly coupled GNSS/IMU factor graph optimization (FGO). This paper formulates ARAIM solution separation and PL computation on the FGO posterior. Because current-epoch pseudorange factors enter only the current-state block of the fixed-lag information matrix, the fault-excluded position covari-ance, solution separation, and nominal-bias propagation follow exactly from the linearized current-state marginal, which re-tains correlations with velocity, IMU-bias, and clock states and with historical information. Woodbury updates with selected-output evaluation compute these quantities for fault-free, single-satellite, constellation-wide, and second-order hypotheses with-out repeated full-state inversion or graph re-optimization. Sur-rounding monitoring layers screen measurements, handle post-fit outliers, and determine PL validity and availability. In two road experiments, both nominal sequences achieve 100% hori-zontal and vertical empirical bounding over PL-valid epochs, with a maximum error-to-PL ratio of 0.96. Normalized errors show that the posterior covariance alone does not capture slowly varying ranging biases: a covariance-only bound computed from the same posteriors retains no resolvable margin, and strong temporal correlation limits the empirical evidence to consisten-cy checks rather than validation of the integrity budget. On identical Experiment A posteriors, the proposed method reduces fault-hypothesis evaluation and PL solving time by 65.2% rela-tive to direct full-state inversion with matching integrity deci-sions, and controlled fault injections demonstrate fail-safe PL invalidation while navigation continues.
https://arxiv.org/abs/2610.32259
Recent robot manipulation policies commonly predict continuous action chunks directly from observations, without explicitly modeling reusable structure in the action space. However, manipulation demonstrations often contain recurring short-horizon motion patterns that are shared across tasks. We introduce \textsc{RoboCode}, which learns these patterns through a shared soft-gated primitive action interface. Rather than assigning each action chunk to a single discrete code, \textsc{RoboCode} represents it as a soft composition of multiple learned primitives, providing a flexible intermediate representation of short-horizon actions. The primitive interface is learned from action chunks pooled across multiple tasks and then frozen while task-specific state encoders are trained to map current state inputs into the shared latent space. Experiments on five ManiSkill manipulation tasks address three questions. First, the learned primitive interface reconstructs action chunks with an average reconstruction MSE of 0.0187. Second, analysis of code activations reveals structured code reuse across tasks. Third, \textsc{RoboCode} achieves the highest or tied closed-loop success rate on all evaluated tasks, with an average success rate of 75.6%, compared with 68.8% for ACT and 41.6% for the larger Diffusion Policy baseline. These results suggest that soft-gated primitive codebooks provide an effective and interpretable intermediate action representation for robot manipulation.
https://arxiv.org/abs/2610.32225
Contact angle (CA) is a common parameter measured in the development of novel contact lens materials and surface coating technologies. Existing literature on in vitro CA of contact lenses via the sessile drop method is hindered in repeatability, reliability, and comparability due to inter-laboratory and inter-operator protocol variation. In this work, the Robotic Autonomous Imaging Surface Evaluator for vision applications (RAISE-Vision) is introduced as an automated platform for CA measurement on curved hydrogel surfaces, with specific application to in vitro contact lens wettability analysis throughout air exposure. Customized 3D printed hardware, such as lens mounts, camera holders and drying apparatuses, was designed to facilitate an automated repeated sessile drop workflow within a liquid handler. Collected images were directly fed to a custom open-source image processing script for CA determination on a curved surface that combines the elliptical fit and tangent methods to allow for variations in droplet shape. The CA of senofilcon A contact lenses measured using the RAISE-Vision platform increased from $62.6 \pm 5.5^\circ$ to $83.4 \pm 2.1^\circ$ over 10 sequential measurements in air and was able to replicate results obtained using an analogous manual protocol ($p = 0.470$). The custom image processing script was also separately demonstrated to generate comparable results to a commercially available CA determination software ($p = 0.343$). Aside from minimizing operator variability for CA measurement, RAISE-Vision provides a platform for standardized in vitro contact lens wettability analysis and understanding of surface wetting behaviour during air exposure.
https://arxiv.org/abs/2610.32188
Robot-assisted dressing is particularly challenging due to narrow openings and articulated body parts. Even small errors in the relative pose between the body part and garment opening can cause snagging, missed insertion, or incomplete dressing. To address these challenges, we formulate dressing as a geometry-aware insertion problem that represents task-relevant spatial relationships between the body part and object opening, rather than relying on the full deformable-object state. Based on this formulation, we propose a Reinforcement Learning (RL) framework combining a staged curriculum, which progressively guides the policy from coarse alignment to insertion and final placement, with a finger-wise insertion soft gate that provides continuous feedback on individual finger insertion progress and helps suppress premature insertion. We evaluate the framework on rigid and deformable bracelet dressing tasks through component ablations, systematic variations in the initial hand pose, and closed-loop execution of the learned policies on a real robotic system. On the post-training deformable evaluation, the full method reaches 84.7 +- 1.2% success. The insertion gate accounts for the largest gain, and adding the curriculum on top of that gate reaches the top score. On a mannequin hand the same full policy succeeds on 15/20 real-robot trials.
https://arxiv.org/abs/2610.32143
Traversability analysis in off-road settings remains a fundamental challenge for mobile robots. Key difficulties include constructing an accurate and expressive local map from multimodal sensor data, and using the map to design traversability rules that yield desirable navigation behavior. Importantly, this system must be resilient to the limited sensing regime brought about by complex environments and high speeds. In this paper, we present TravSUITE, a traversability system suitable for high-speed navigation in off-road environments. TravSUITE consists of two major components: 1) an efficient voxel mapper that leverages visual foundation models (VFMs) to build a rich geometric-semantic local map from streams of on-board sensor data, and 2) a unified neural network that jointly predicts traversability-relevant quantities in bird's eye view (BEV), including geometry, semantics, speed and cost. Our training strategy is entirely annotation-free and self-supervised, leveraging tasks such as map inpainting and inverse reinforcement learning (IRL) to learn both map representations and traversability. We perform a thorough ablation study and comparison to state-of-the-art approaches, and the results indicate that cost learning and auxiliary inpainting each contribute significantly to planning quality, and their combination is critical for achieving state-of-the-art performance in off-road path planning. We also design a simple risk adaptation mechanism to leverage our method's uncertainty estimates at deploy-time, and demonstrate that a combination of inpainting and risk estimation can result in 80% fewer navigation errors and 5% faster autonomous traversal speeds in real-world hardware experiments.
https://arxiv.org/abs/2610.32130
Flow-matching Vision-Language-Action (VLA) models generate expressive continuous trajectories for robotic manipulation, yet balancing real-time responsiveness with execution stability remains a persistent challenge. Standard deployment workflows typically treat numerical solver depth and execution horizon as decoupled choices. The horizon amortizes inference latency, while solver depth is kept high to preserve integration fidelity. In this paper, we reveal a fundamental compute-execution coupling in flow-matching VLAs governed by a systematic sign reversal. While coarse numerical integration degrades precision under rapid replanning, it markedly improves task performance under extended open-loop commitment. Geometrically, numerical truncation contracts lateral action dispersion around an invariant trajectory centroid, serving as an implicit regularizer that suppresses compounding open-loop drift. Across LIBERO, RoboTwin 2.0, and real-world Franka manipulation on representative VLA backbones including $\pi_0$, $\pi_{0.5}$, and GR00T, exploiting this joint coupling consistently achieves superior task success while drastically lowering inference latency, entirely training-free and without architectural modifications. These findings establish numerical sampling depth as an operational variance regulator, shifting its role from a static precision budget to an active deployment knob in robot learning. Project page: this https URL
https://arxiv.org/abs/2610.32102
Behavior cloning for robotic manipulation is highly sensitive to the coverage of its training data, yet collecting diverse successful demonstrations remains costly. We study reinforcement learning (RL) not as a deployable controller, but as a data generator for downstream visuomotor imitation learning. Our key premise is that multiple successful solution modes cover a broader range of successful trajectories than repeatedly collecting a single narrow solution. We introduce DiscoDemo, which trains a privileged latent-conditioned RL generator with a parallel reverse curriculum and a success-gated diversity objective to discover diverse solutions while preserving task success. The trained generator produces multiple successful closed-loop behaviors, which are safety-filtered and exported as a static dataset for downstream imitation learning. Across four manipulation tasks, DiscoDemo produces broader successful-state coverage and improves downstream visuomotor policies in both simulation and the real world. Averaged across tasks, downstream success increases to 62% from 40% for the strongest baseline in simulation, and to 78% from 39% on the real robot, respectively. These results suggest that successful solution diversity improves imitation learning for a pretrained flow-matching VLA and remains complementary to corrective data collection. Project page: this https URL
https://arxiv.org/abs/2610.32093
Contact-rich manipulation underlies many everyday human activities, yet evaluating robots on such tasks remains difficult: the tasks involve diverse contact behaviors and are hard to reproduce across laboratories. Drawing on observations of everyday human manipulation, we identify eight essential skills that provide a practical framework for studying contact-rich manipulation. We introduce ContactBench 1.0, an extensible benchmark that instantiates these skills through 36 reference tasks and more than 40 object assets. The assets capture task-relevant geometry and contact behavior, can be fabricated reproducibly on consumer-grade 3D printers, and are each paired with a matching simulation model. Controlled task variations in both the real world and simulation enable evaluation of how well robot systems generalize. We evaluate recent representative vision-only and visual-tactile policy learning methods on six benchmark tasks. The results reveal that while vision-only policies struggle to perceive or react to contact events, visual-tactile policies do not meaningfully improve their performance, highlighting that contact-rich manipulation remains a significant challenge for current policy learning methods. By providing common task definitions alongside physical and simulated assets, ContactBench 1.0 offers a shared testbed for developing, evaluating, and comparing contact-rich manipulation systems. The benchmark will be open-sourced at this https URL.
https://arxiv.org/abs/2610.24512
Robotic minimally invasive surgery offers well-documented clinical benefits, but the cost and infrastructure requirements of purpose-built platforms limit access in rural and lower-resourced facilities. Recent work has teleoperated general-purpose robots for laparoscopic tasks and in vivo procedures, but relied on handheld instruments coupled through passive linkages rather than native robotic actuation. Instead, we adapt da Vinci Classic and Xi instruments onto general-purpose robots that can integrate in clinical workflows. We present DASH, a da Vinci Adapter for Serial-link and Humanoid platforms, consisting of two types of adapters that require no modification to the instruments themselves. Each adapter is compatible with many robotic platform load capacities, engages the instrument's native latch, and wirelessly identifies inserted tools to load instrument-specific kinematics and coupling matrices. Teleoperated experiments demonstrate the efficacy of DASH and its benefits for enabling research access to surgical robotic platforms.
https://arxiv.org/abs/2610.05792
Physical self-reproduction in modular robots remains difficult because a robot must not only move, but also collect, align, dock, and separate modules reliably in hardware. Taking inspiration from how genetically identical cells form the varied tissues of an organism, we introduce a single standardized robotic module that combines a magnetic docking interface with two actuated joints and folds from a flat, two-dimensional configuration into three-dimensional structures. Modules of this single design can be assembled into two morphologies: a three-module "trilobite" and a five-module "seal". Using a small set of discrete motion commands, a human operator steers an assembled parent robot to dock magnetically with a sequence of spare modules, which are assembled into a second, structurally identical individual that then separates from the parent. Because the parent retains all of its original modules, this is a budding-style form of physical reproduction. Here, we report preliminary hardware demonstrations of this full process on both morphologies. We discuss the current scope of the system, including its present reliance on direct human teleoperation rather than autonomous control, and outline directions toward autonomous execution of this process and repeated validation across a larger number of independent trials.
https://arxiv.org/abs/2610.31929
As humanoid robots move from research laboratories toward deployment in human-centered environments such as homes, hospitals, and warehouses, safe operation around nearby people and objects becomes a fundamental requirement. Safety filters have emerged as a promising technique to address this challenge by providing mathematical guarantees that generated control actions do not result in collisions. This paper shows that these guarantees hold only when the safety parameters are configured correctly. However, correct parameter configurations are task- and environment-specific, and no general method exists to determine them across diverse deployments. As a result, safety filters shipped with humanoid robots can fail after deployment in specific scenarios. We show that attackers can exploit these failures through only normal user interactions without any additional privileges or capabilities. To demonstrate this risk, we present goal insertion attacks and develop SIREN, an automated tool that identifies vulnerable inserting goals for a given task and environment. Experimental results show that SIREN identifies attacks in all vulnerable scenarios in the SPARK dataset. Compared with a baseline approach, SIREN discovers three times as many attacks. We also show that SIREN is able to generate attacks against a real humanoid and cause it to collide while executing tasks. These results demonstrate that current safety filters introduce exploitable vulnerabilities that can cause safety violations, which reveals that existing safety mechanisms can themselves become a source of safety risk for humanoid robots.
https://arxiv.org/abs/2610.31923
In traditional Gaussian process (GP) regression, the inputs are typically assumed to be known, and uncertainty enters only through the measurements. When scalar fields are mapped using GP regression with measurements obtained from a mobile sensor, the inputs correspond to the estimated pose of the mobile sensor. The goal in this letter is to understand how a recorded GP, trained on erroneous inputs (e.g., pose estimates from dead-reckoning), can be corrected in a numerically efficient way when better estimates of the inputs become available. In the developed bias-aware GP (Ba-GP) framework, the recorded GP posterior-mean gradient is used to propagate estimated input bias and noise into mean and covariance corrections. The resulting additive covariance structure is then exploited to develop numerically efficient first-order correction algorithms. A simulation study and an indoor light-intensity mapping experiment show that Ba-GP improves predictive accuracy and uncertainty calibration relative to noisy-input GP baselines, with an order-of-magnitude reduction in computational cost relative to full GP retraining.
https://arxiv.org/abs/2610.31920
Swarm morphogenesis requires robots to assemble desired shapes through local interactions. We present a modular framework integrating neural cellular automata (NCA) with existing site selection and collision-aware control. Within this framework, we introduce a decentralized attachment-site discovery mechanism. Robots already in the formation update and share local NCA states to identify new attachment sites, enabling progressive growth without a predefined list of sites during deployment. To bridge the mismatch between conventional NCA growth and gradual physical assembly, we additionally propose offline sparse-retention fine-tuning that simulates partial attachment, together with a false-positive loss that penalizes invalid site activations. Simulations demonstrate formation of diverse lattice morphologies, scaling to several hundred robots, and substantial improvements in morphology accuracy from the proposed fine-tuning. Experiments with Crazyflie quadrotors validate the integrated execution pipeline using robot-relative observations reconstructed from external tracking.
https://arxiv.org/abs/2610.31900
Production lane keeping assist (LKA) systems tell the driver whether they detect the lane markings, but whether this displayed status reflects what the system perceives, and what happens when it does not, has not been examined in real-world driving. This study introduces CANDID-LKA, a CAN-based Diagnosis of the Indicated Detection status in LKA. It evaluates a proprietary driver-facing status using three signals: the displayed status and the steering torque of the driver, both recorded from the controller area network (CAN) bus of the vehicle, and the lane perception confidence of a third-party vision system observing the same road. The framework examines how the status responds to driving conditions and estimates how completely it reports degraded lane perception by comparing it with the other two signals in a latent-class model. We apply it to 39,800 seconds of real-world driving from 13 vehicle models of three manufacturers. The results show that the status does not respond consistently to the conditions that make markings harder to see, and that the three makes differ widely in how often they report not detected. The latent-class model shows that the displayed status is incomplete: it reports only 47% of degraded seconds as not detected, compared with 80% for the third-party low perception confidence, and in half of the low-confidence episodes it never reports not detected. In the silent seconds in which the status reads detected while the perception confidence is low and the driver is not in control, the standard deviation of lateral position over the next 3 s is about 60% larger than when the status and the perception confidence agree. A detected status thus does not guarantee reliable lane perception, and silence with the driver not in control is the case that most needs attention.
https://arxiv.org/abs/2610.31899
Data-driven robots must learn to avoid collisions with as few demonstrations as possible to avoid collecting large datasets of destructive collision examples. We explore whole-body proximity sensors as inductive biases for learning reflexive collision-dodging behavior and find that lossy sensors with low resolution were stronger biases than high-resolution coverage. We also found that, perhaps unintuitively, anticipating when and where contact happens on the body of a robot reduced its learning sample efficiency. We investigated eight candidate representations in a reinforcement learning environment, which included latent compression, future contact anticipation, lossy corruption, and privileged state estimation. The candidate representations were tested by training over 200 humanoid robots to play a simplified version of dodgeball end-to-end under an intentionally constrained sample and compute budget.
https://arxiv.org/abs/2610.31891
Swarms of small robots are an attractive way to survey lunar caves and lava tubes. Missions such as NASA's CADRE will fly cooperating rovers for distributed mapping, but only in daylight. The 354 h lunar night drops the surface towards 95 K, and the radioisotope heaters that let single rovers survive it cannot be supplied at swarm scale. In this study we investigate whether a swarm can instead use its own members as thermal shielding. We train one policy, shared by all twenty robots, by reinforcement learning with an evolution-strategies estimator. The policy receives a single reward, once, at the end of the night: the mean temperature of the five coldest robots. Nothing in that reward refers to position, spacing, exposure or motion. Two behaviors nonetheless emerge. The swarm aggregates, and robots circulate between the sheltered core and the exposed rim, the same pair of behaviors with which emperor penguins survive the Antarctic winter. We measure the survival threshold, the smallest stored energy per robot at which every robot is still alive at sunrise. Across ten runs it is 155.6 +/- 1.3 pcm-h, where one pcm-h is 7.2 kJ of stored heat. That is 37% below a motionless huddle and within 13% of a physics-based lower bound, and at a budget where the motionless huddle loses six robots, every robot survives. The behavior needs no shared spatial reference and no robot-robot contact, and transfers from twenty robots to forty without retraining.
https://arxiv.org/abs/2610.31868
When a vision-language-action (VLA) model is fine-tuned with its trainable weights in bfloat16 (bf16) and no fp32 master copy, AdamW writes each update straight into bf16, silently rounding away any update smaller than 1/512 of a weight. In 9 of 12 open VLA codebases a default or documented recipe trains this way, 4 on full weight matrices, including LeRobot's SmolVLA default and its $\pi_0$/$\pi_{0.5}$ guides. In our LIBERO runs without the copy, a probe every 100 steps finds at least 79% of updates to bf16 weights erased. On the Hugging Face Hub, 99% of sampled full-weight SmolVLA fine-tunes store the action expert in bf16, and 84% of recent $\pi_{0.5}$ fine-tunes with a training configuration set the bf16 flag. On LIBERO-Object, with rules fixed in advance, an fp32 copy raises success over LeRobot's storage from 4.3% to 49.3% for SmolVLA and from 5.1% to 80.9% for $\pi_{0.5}$ at a ten-times lower learning rate, where both models are undertrained, and from 89.0% to 97.7% for $\pi_{0.5}$ at its default rate. For SmolVLA at its default rate it adds 10.2 points with open-loop 50-action chunks but only 4.2 when re-planning every 10 actions. In these five comparisons and two more against SmolVLA with every trainable weight in bf16, every run with the copy beats every run without it (p < 0.05); all gaps but the 8.7- and 4.2-point ones clear our 10-point threshold. The fix is to keep the trainable weights in fp32 during training, a one-line change for SmolVLA in LeRobot.
https://arxiv.org/abs/2610.31859