Retrieve, Don't Retrain: Extending Vision Language Action Models to New Tasks at Test Time

Extending a vision-language-action (VLA) policy to a new task typically requires task-specific teleoperated demonstrations and per-task fine-tuning, making adaptation costly in both data collection and compute. In this paper, we show that this target-side per-task adaptation cost can be replaced by retrieval. Our retrieval-augmented policy is trained once on paired demonstrations from the target embodiment (query) and a cheaper embodiment (pool, e.g., human-hand video), then frozen. New tasks are added at deployment by appending pool-side demonstrations to a retrieval pool. The frozen policy conditions on retrieved trajectories at every control step, so new tasks are absorbed by indexing data rather than updating parameters. Fine-tuning is needed only to take on a new, unseen embodiment, not for each new task. We show that retrieval improves policies beyond a specific backbone, including standard VLA policies, but its effect is especially pronounced in Cosmos Policy, a video-generation-based world-action model (WAM). In this setting, retrieval supplies coarse task progression, while the WAM's future-image objective provides an additional visual consistency signal that strengthens the retrieval-conditioned actions. On PushT, we study how retrieval provides a reusable high-level motion prior for cross-embodiment generalization to unseen goal angles, while on RoboTwin 2.0 our method outperforms cross-embodiment baselines on unseen tasks, and we additionally demonstrate the method on a real robot.

Reinforcement Learning for Risk Adaptation via Differentiable CVaR Barrier Functions

Planning through crowded environments under uncertain obstacle motions remains difficult, as stochastic interactions often induce overly conservative behavior or reduced efficiency. To address this challenge, we propose an end-to-end risk adaptation framework for crowd navigation under obstacle-motion uncertainty modeled by a Gaussian mixture model. The framework combines reinforcement learning~(RL) with a differentiable quadratic-program safety layer based on Conditional Value-at-Risk~(CVaR) barrier functions, jointly learning nominal control input, risk level, and safety margin and enforcing explicit probabilistic safety constraints. This design enables context-aware adaptation, promoting efficient behavior while invoking caution only when necessary. We conduct extensive evaluations in dynamic, uncertain, and crowded environments across varying obstacle densities and robot models, and further assess generalization under three out-of-distribution cases. Comparisons across optimization-based, RL-based, and integrated RL and optimization methods are provided, and the proposed method is shown to deliver the strongest overall performance in safety, efficiency, and generalization under uncertainty.

Distributionally Robust Safety Under Arbitrary Uncertainties: A Safety Filtering Approach

In this work, we study how to ensure probabilistic safety for nonlinear systems under distributional ambiguity. Our approach builds on a backup-based safety filtering framework that switches between a high-performance nominal policy and a certified backup policy to ensure safety. To handle arbitrary uncertainties from ambiguous distributions, i.e., where the distribution is not of specific structure and the true distribution is unknown, we adopt a distributionally robust (DR) formulation using Wasserstein ambiguity sets. Rather than solving a high-dimensional DR trajectory optimization problem online, we exploit the structure of backup-based safety filtering to reduce safety certification to a one-dimensional search over the switching time between nominal and backup policies. We then develop a sampling-based certification procedure with finite-sample guarantees, where empirical failure probabilities are compared against a Wasserstein-inflated threshold. We validate our method through simulations across three systems, from a Dubins vehicle to a high-speed racing car and a fighter jet, demonstrating the broad applicability and computational efficiency.

Backup-Based Safety Filters: A Comparative Review of Backup CBF, Model Predictive Shielding, and gatekeeper

This paper revisits three backup-based safety filters -- Backup Control Barrier Functions (Backup CBF), Model Predictive Shielding (MPS), and gatekeeper -- through a unified comparative framework. Using a common safety-filter abstraction and shared notation, we make explicit both their common backup-policy structure and their key algorithmic differences. We compare the three methods through their filter-inactive sets, i.e., the states where the nominal policy is left unchanged. In particular, we show that MPS is a special case of gatekeeper, and we further relate gatekeeper to the interior of the Backup CBF inactive set within the implicit safe set. This unified view also highlights a key source of conservatism in backup-based safety filters: safety is often evaluated through the feasibility of a backup maneuver, rather than through the nominal policy's continued safe execution. The paper is intended as a compact tutorial and review that clarifies the theoretical connections and differences among these methods.

MuCo: Multi-turn Contrastive Learning for Multimodal Embedding Model

Universal Multimodal embedding models built on Multimodal Large Language Models (MLLMs) have traditionally employed contrastive learning, which aligns representations of query-target pairs across different modalities. Yet, despite its empirical success, they are primarily built on a "single-turn" formulation where each query-target pair is treated as an independent data point. This paradigm leads to computational inefficiency when scaling, as it requires a separate forward pass for each pair and overlooks potential contextual relationships between multiple queries that can relate to the same context. In this work, we introduce Multi-Turn Contrastive Learning (MuCo), a dialogue-inspired framework that revisits this process. MuCo leverages the conversational nature of MLLMs to process multiple, related query-target pairs associated with a single image within a single forward pass. This allows us to extract a set of multiple query and target embeddings simultaneously, conditioned on a shared context representation, amplifying the effective batch size and overall training efficiency. Experiments exhibit MuCo with a newly curated 5M multimodal multi-turn dataset (M3T), which yields state-of-the-art retrieval performance on MMEB and M-BEIR benchmarks, while markedly enhancing both training efficiency and representation coherence across modalities. Code and M3T are available at https://github.com/naver-ai/muco

Beyond Collision Cones: Dynamic Obstacle Avoidance for Nonholonomic Robots via Dynamic Parabolic Control Barrier Functions

Control Barrier Functions (CBFs) are a powerful tool for ensuring the safety of autonomous systems, yet applying them to nonholonomic robots in cluttered, dynamic environments remains an open challenge. State-of-the-art methods often rely on collision-cone or velocity-obstacle constraints which, by only considering the angle of the relative velocity, are inherently conservative and can render the CBF-based quadratic program infeasible, particularly in dense scenarios. To address this issue, we propose a Dynamic Parabolic Control Barrier Function (DPCBF) that defines the safe set using a parabolic boundary. The parabola's vertex and curvature dynamically adapt based on both the distance to an obstacle and the magnitude of the relative velocity, creating a less restrictive safety constraint. We prove that the proposed DPCBF is valid for a kinematic bicycle model subject to input constraints. Extensive comparative simulations demonstrate that our DPCBF-based controller significantly enhances navigation success rates and QP feasibility compared to baseline methods. Our approach successfully navigates through dense environments with up to 100 dynamic obstacles, scenarios where collision cone-based methods fail due to infeasibility.

Token Bottleneck: One Token to Remember Dynamics

Deriving compact and temporally aware visual representations from dynamic scenes is essential for successful execution of sequential scene understanding tasks such as visual tracking and robotic manipulation. In this paper, we introduce Token Bottleneck (ToBo), a simple yet intuitive self-supervised learning pipeline that squeezes a scene into a bottleneck token and predicts the subsequent scene using minimal patches as hints. The ToBo pipeline facilitates the learning of sequential scene representations by conservatively encoding the reference scene into a compact bottleneck token during the squeeze step. In the expansion step, we guide the model to capture temporal dynamics by predicting the target scene using the bottleneck token along with few target patches as hints. This design encourages the vision backbone to embed temporal dependencies, thereby enabling understanding of dynamic transitions across scenes. Extensive experiments in diverse sequential tasks, including video label propagation and robot manipulation in simulated environments demonstrate the superiority of ToBo over baselines. Moreover, deploying our pre-trained model on physical robots confirms its robustness and effectiveness in real-world environments. We further validate the scalability of ToBo across different model scales.

BR-MPPI: Barrier Rate guided MPPI for Enforcing Multiple Inequality Constraints with Learned Signed Distance Field

Model Predictive Path Integral (MPPI) controller is used to solve unconstrained optimal control problems and Control Barrier Function (CBF) is a tool to impose strict inequality constraints, a.k.a, barrier constraints. In this work, we propose an integration of these two methods that employ CBF-like conditions to guide the control sampling procedure of MPPI. CBFs provide an inequality constraint restricting the rate of change of barrier functions by a classK function of the barrier itself. We instead impose the CBF condition as an equality constraint by choosing a parametric linear classK function and treating this parameter as a state in an augmented system. The time derivative of this parameter acts as an additional control input that is designed by MPPI. A cost function is further designed to reignite Nagumo's theorem at the boundary of the safe set by promoting specific values of classK parameter to enforce safety. Our problem formulation results in an MPPI subject to multiple state and control-dependent equality constraints which are non-trivial to satisfy with randomly sampled control inputs. We therefore also introduce state transformations and control projection operations, inspired by the literature on path planning for manifolds, to resolve the aforementioned issue. We show empirically through simulations and experiments on quadrotor that our proposed algorithm exhibits better sampled efficiency and enhanced capability to operate closer to the safe set boundary over vanilla MPPI.

Certifiably-Correct Mapping for Safe Navigation Despite Odometry Drift

Accurate perception, state estimation and mapping are essential for safe robotic navigation as planners and controllers rely on these components for safety-critical decisions. However, existing mapping approaches often assume perfect pose estimates, an unrealistic assumption that can lead to incorrect obstacle maps and therefore collisions. This paper introduces a framework for certifiably-correct mapping that ensures that the obstacle map correctly classifies obstacle-free regions despite the odometry drift in vision-based localization systems (VIO}/SLAM). By deflating the safe region based on the incremental odometry error at each timestep, we ensure that the map remains accurate and reliable locally around the robot, even as the overall odometry error with respect to the inertial frame grows unbounded. Our contributions include two approaches to modify popular obstacle mapping paradigms, (I) Safe Flight Corridors, and (II) Signed Distance Fields. We formally prove the correctness of both methods, and describe how they integrate with existing planning and control modules. Simulations using the Replica dataset highlight the efficacy of our methods compared to state-of-the-art techniques. Real-world experiments with a robotic rover show that, while baseline methods result in collisions with previously mapped obstacles, the proposed framework enables the rover to safely stop before potential collisions.

Safe Navigation in Uncertain Crowded Environments Using Risk Adaptive CVaR Barrier Functions

Robot navigation in dynamic, crowded environments poses a significant challenge due to the inherent uncertainties in the obstacle model. In this work, we propose a risk-adaptive approach based on the Conditional Value-at-Risk Barrier Function (CVaR-BF), where the risk level is automatically adjusted to accept the minimum necessary risk, achieving a good performance in terms of safety and optimization feasibility under uncertainty. Additionally, we introduce a dynamic zone-based barrier function which characterizes the collision likelihood by evaluating the relative state between the robot and the obstacle. By integrating risk adaptation with this new function, our approach adaptively expands the safety margin, enabling the robot to proactively avoid obstacles in highly dynamic environments. Comparisons and ablation studies demonstrate that our method outperforms existing social navigation approaches, and validate the effectiveness of our proposed framework.