Recursive Belief Vision Language Action Models

Vision-language-action models must enable agents to execute long-horizon tasks under partial observability. However, most existing approaches remain observation-driven, relying on short context windows or repeated queries to vision-language models (VLMs). This leads to loss of task progress, action repetition under perceptual aliasing, and high inference latency. While semantic grounding is important, long-horizon manipulation fundamentally requires persistent, action-conditioned state representations. Current VLAs lack such representations and exhibit limited temporal and physical reasoning, making them ill-suited for multi-stage control. This paper introduces RB-VLA, a belief-centric architecture trained with self-supervised world-model objectives that maintains a compact latent state encoding task-relevant history, dynamics, and object interactions. Queried once per task, the VLM provides high-level intent, while the belief tracks task progress and enables phase-aware, causally grounded control under partial observability without storing raw observations or scaling memory with time. The belief and intent jointly condition a diffusion policy for robust closed-loop execution. RB-VLA outperforms prior VLAs on long-horizon benchmarks, achieving 52.5 percent and 37.5 percent higher success rates on multi-stage pick-and-place and stacking tasks, respectively, compared to pi_0. It also reduces inference latency by up to five times relative to baselines and eliminates memory growth across timesteps observed in existing VLAs. Ablations show the belief module is the primary driver of performance, increasing success rates from 32.5 percent without belief to 77.5 percent with belief.

ClutterNav: Gradient-Guided Search for Efficient 3D Clutter Removal with Learned Costmaps

Dense clutter removal for target object retrieval presents a challenging problem, especially when targets are embedded deep within densely-packed configurations. It requires foresight to minimize overall changes to the clutter configuration while accessing target objects, avoiding stack destabilization and reducing the number of object removals required. Rule-based planners when applied to this problem, rely on rigid heuristics, leading to high computational overhead. End-to-end reinforcement learning approaches struggle with interpretability and generalizability over different conditions. To address these issues, we present ClutterNav, a novel decision-making framework that can identify the next best object to be removed so as to access a target object in a given clutter, while minimising stack disturbances. ClutterNav formulates the problem as a continuous reinforcement learning task, where each object removal dynamically updates the understanding of the scene. A removability critic, trained from demonstrations, estimates the cost of removing any given object based on geometric and spatial features. This learned cost is complemented by integrated gradients that assess how the presence or removal of surrounding objects influences the accessibility of the target. By dynamically prioritizing actions that balance immediate removability against long-term target exposure, ClutterNav achieves near human-like strategic sequencing, without predefined heuristics. The proposed approach is validated extensively in simulation and over real-world experiments. The results demonstrate real-time, occlusion-aware decision-making in partially observable environments.

QuickGrasp: Lightweight Antipodal Grasp Planning with Point Clouds

Grasping has been a long-standing challenge in facilitating the final interface between a robot and the environment. As environments and tasks become complicated, the need to embed higher intelligence to infer from the surroundings and act on them has become necessary. Although most methods utilize techniques to estimate grasp pose by treating the problem via pure sampling-based approaches in the six-degree-of-freedom space or as a learning problem, they usually fail in real-life settings owing to poor generalization across domains. In addition, the time taken to generate the grasp plan and the lack of repeatability, owing to sampling inefficiency and the probabilistic nature of existing grasp planning approaches, severely limits their application in real-world tasks. This paper presents a lightweight analytical approach towards robotic grasp planning, particularly antipodal grasps, with little to no sampling in the six-degree-of-freedom space. The proposed grasp planning algorithm is formulated as an optimization problem towards estimating grasp points on the object surface instead of directly estimating the end-effector pose. To this extent, a soft-region-growing algorithm is presented for effective plane segmentation, even in the case of curved surfaces. An optimization-based quality metric is then used for the evaluation of grasp points to ensure indirect force closure. The proposed grasp framework is compared with the existing state-of-the-art grasp planning approach, Grasp pose detection (GPD), as a baseline over multiple simulated objects. The effectiveness of the proposed approach in comparison to GPD is also evaluated in a real-world setting using image and point-cloud data, with the planned grasps being executed using a ROBOTIQ gripper and UR5 manipulator.