Vision-TL-Action: Neuro-Symbolic Trajectory Generation from Visual Observations and Temporal Logic

Temporal logic (TL) provides a compositional language for the formulation of long horizon robotic tasks, but existing TL-conditioned trajectory generators can sidestep perception-to-symbol binding by encoding exact object geometry in the task graph. We introduce \emph{Vision-TL-Action}, which generates action trajectories from multi-view images, a coordinate-free TL syntax graph, and the robot initial state. TL-node tokens and spatial visual tokens are fused through bidirectional cross-attention, and the resulting representation conditions a flow-matching trajectory generator. Visual tokens are augmented only with normalized image-plane locations and camera-view identifiers, while a training-only predicate-to-region objective encourages grounding to referenced objects. Consistent with prior work in this domain, we evaluate the model using Success@$K$, the fraction of tasks for which at least one of K sampled trajectories satisfies the TL specification. On Panda task, our model achieves 67.45% Success@1024, compared with 59.11% for the oracle-state baseline. On AntMaze task, it achieves 96.35% Success@256, comparable to the oracle result of 96.88%. Resolution and intervention studies show that spatial detail depends on semantic grounding and predicate identity affects both attention and performance. These results demonstrate a direct mapping from visual observations and structured TL goals to action trajectories without requiring object geometry at inference. Code is available at https://github.com/AricLau07/vision-tl-action.

VLA-JEPA: Enhancing Vision-Language-Action Model with Latent World Model

Pretraining Vision-Language-Action (VLA) policies on internet-scale video is appealing, yet current latent-action objectives often learn the wrong thing: they remain anchored to pixel variation rather than action-relevant state transitions, making them vulnerable to appearance bias, nuisance motion, and information leakage. We introduce VLA-JEPA, a JEPA-style pretraining framework that sidesteps these pitfalls by design. The key idea is leakage-free state prediction: a target encoder produces latent representations from future frames, while the student pathway sees only the current observation -- future information is used solely as supervision targets, never as input. By predicting in latent space rather than pixel space, VLA-JEPA learns dynamics abstractions that are robust to camera motion and irrelevant background changes. This yields a simple two-stage recipe -- JEPA pretraining followed by action-head fine-tuning -- without the multi-stage complexity of prior latent-action pipelines. Experiments on LIBERO, LIBERO-Plus, SimplerEnv and real-world manipulation tasks show that VLA-JEPA achieves consistent gains in generalization and robustness over existing methods.

A General One-Shot Multimodal Active Perception Framework for Robotic Manipulation: Learning to Predict Optimal Viewpoint

Active perception in vision-based robotic manipulation aims to move the camera toward more informative observation viewpoints, thereby providing high-quality perceptual inputs for downstream tasks. Most existing active perception methods rely on iterative optimization, leading to high time and motion costs, and are tightly coupled with task-specific objectives, which limits their transferability. In this paper, we propose a general one-shot multimodal active perception framework for robotic manipulation. The framework enables direct inference of optimal viewpoints and comprises a data collection pipeline and an optimal viewpoint prediction network. Specifically, the framework decouples viewpoint quality evaluation from the overall architecture, supporting heterogeneous task requirements. Optimal viewpoints are defined through systematic sampling and evaluation of candidate viewpoints, after which large-scale training datasets are constructed via domain randomization. Moreover, a multimodal optimal viewpoint prediction network is developed, leveraging cross-attention to align and fuse multimodal features and directly predict camera pose adjustments. The proposed framework is instantiated in robotic grasping under viewpoint-constrained environments. Experimental results demonstrate that active perception guided by the framework significantly improves grasp success rates. Notably, real-world evaluations achieve nearly double the grasp success rate and enable seamless sim-to-real transfer without additional fine-tuning, demonstrating the effectiveness of the proposed framework.