Q. Are relationships between individual instances made explicit (e.g., users' movie ratings,

Vision Language Models (VLMs) are increasingly being used in a broad range of applications, bringing their security and behavioral control to the forefront. While existing approaches for behavioral control or output redirection, like system prompting in VLMs, are easily detectable and often ineffective, activation-based steering vectors require invasive runtime access to model internals--incompatible with API-based services and closed-source deployments. We introduce VISOR (Visual Input-based Steering for Output Redirection), a novel method that achieves sophisticated behavioral control through optimized visual inputs alone. By crafting universal steering images that induce target activation patterns, VISOR enables practical deployment across all VLM serving modalities while remaining imperceptible compared to explicit textual instructions. We validate VISOR on LLaVA-1.5-7B across three critical alignment tasks: refusal, sycophancy and survival instinct. A single 150KB steering image matches steering vector performance within 1-2% for positive behavioral shifts while dramatically exceeding it for negative steering--achieving up to 25% shifts from baseline compared to steering vectors' modest changes. Unlike system prompting (3-4% shifts), VISOR provides robust bidirectional control while maintaining 99.9% performance on 14,000 unrelated MMLU tasks. Beyond eliminating runtime overhead and model access requirements, VISOR exposes a critical security vulnerability: adversaries can achieve sophisticated behavioral manipulation through visual channels alone, bypassing text-based defenses. Our work fundamentally re-imagines multimodal model control and highlights the urgent need for defenses against visual steering attacks.

Abstract-- Dexterous robotic manipulator teleoperation is widely used in many applications, either where it is convenient to keep the human inside the control loop, or to train advanced robot agents. So far, this technology has been used in combination with camera systems with remarkable success. On the other hand, only a limited number of studies have focused on leveraging haptic feedback from tactile sensors in contexts where camera-based systems fail, such as due to self-occlusions or poor light conditions like smoke. This study demonstrates the feasibility of precise pick-and-place teleoperation without cameras by leveraging tactile-based 3D object reconstruction in VR and providing haptic feedback to a blindfolded user. Our preliminary results show that integrating these technologies enables the successful completion of telemanipulation tasks previously dependent on cameras, paving the way for more complex future applications.

The Immersed Visor aims for spatial computing's sweet spot The $1,050 device has 4K per-eye resolution and weighs less than an iPhone 16 Pro. An Austin-based startup best known for its VR and mixed reality workspace software for other companies' headsets now has hardware of its own. The Immersed Visor appears to sit somewhere between a Vision Pro Lite and Xreal Plus: a lightweight head-worn device that creates a high-resolution spatial computing environment on the cheap (well, relatively speaking). Teased to death for months, Immersed founder Renji Bijoy finally unveiled the Visor at an Austin event on Thursday. The device, a bit more than glasses but much less than a full headset, gives each eye the equivalent of a 4K OLED screen.

Spatial understanding is a fundamental aspect of computer vision and integral for human-level reasoning about images, making it an important component for grounded language understanding. While recent text-to-image synthesis (T2I) models have shown unprecedented improvements in photorealism, it is unclear whether they have reliable spatial understanding capabilities. We investigate the ability of T2I models to generate correct spatial relationships among objects and present VISOR, an evaluation metric that captures how accurately the spatial relationship described in text is generated in the image. To benchmark existing models, we introduce a dataset, $\mathrm{SR}_{2D}$, that contains sentences describing two or more objects and the spatial relationships between them. We construct an automated evaluation pipeline to recognize objects and their spatial relationships, and employ it in a large-scale evaluation of T2I models. Our experiments reveal a surprising finding that, although state-of-the-art T2I models exhibit high image quality, they are severely limited in their ability to generate multiple objects or the specified spatial relations between them. Our analyses demonstrate several biases and artifacts of T2I models such as the difficulty with generating multiple objects, a bias towards generating the first object mentioned, spatially inconsistent outputs for equivalent relationships, and a correlation between object co-occurrence and spatial understanding capabilities. We conduct a human study that shows the alignment between VISOR and human judgement about spatial understanding. We offer the $\mathrm{SR}_{2D}$ dataset and the VISOR metric to the community in support of T2I reasoning research.