VisualClaw: A Real-Time, Personalized Agent for the Physical World

2026-06-15

AI Summary

This work addresses the challenges of deploying vision-language models in the physical world—namely high latency, excessive cost, static architectures, and the absence of tool-use evaluation benchmarks—by introducing a self-evolving multimodal agent. The proposed approach integrates cascaded gated frame filtering with a hot-cold skill injection mechanism to reduce video processing overhead through hybrid encoding and dynamically updates its skill repertoire via memory retrieval for efficient, personalized interaction. Furthermore, the authors introduce VisualClawArena, a novel benchmark emphasizing visual grounding and dynamic execution. Experiments demonstrate that the method reduces API costs by up to 98% on average across multiple video question-answering benchmarks while achieving accuracy gains of up to 15.80%; on VisualClawArena, it improves macro accuracy by 2.9%–3.2% and lowers computational cost by 9.5%.

Abstract

Vision language models are serving as general-purpose interfaces for complex multimodal tasks. However, deployment still faces three gaps: VLMs typically incur high latency and cost when processing dense video frames and long prompts, the agent scaffold remains static after deployment, and standard video-QA benchmarks do not test whether agents can use visual evidence inside tool-using workspaces. We present VisualClaw, a self-evolving multimodal agent built around two principles. First, hybrid encoding reduces deployment cost by filtering less informative streaming frames with a cascaded gate and compressing the text skill bank through hot/cold top-k injection. Second, skill evolution lets the agent learn from failures: retrieved memories condition an evolver as direct concatenated context or as guided evidence, producing skill-bank updates that help future questions. Across 4 video-QA benchmarks with 2 VLMs, VisualClaw cuts per-question API cost by an average -98% versus full-frame upload and by -25.9% over the offline uniform 8 frame baseline, while boosting accuracy in most settings, e.g., an average +3.85% and a peak +15.80% on EgoSchema with Gemini 3 Flash. To address the gap, we curate VisualClawArena, a 200-scenario multimodal agentic benchmark built through a strict five-stage pipeline; models must use video evidence, documents, dynamic updates, and executable checks inside a workspace. On VisualClawArena, the same framework with computer-use agent backends improves macro accuracy by +2.9% for Codex (GPT-5.5) and +3.2% for Claude Code (Sonnet 4.6) over no-evolution baselines, with a -9.5% cost reduction compared to the uniform-sampled baseline. These properties make VisualClaw a natural fit for edge applications, where the cascade reduces a 1-hour streaming session from ~3,600 API uploads down to only 5-20 calls and the self-evolution makes it a perfect personalized assistant.

Problem

• vision-language models
• multimodal agent
• video-QA benchmarks
• deployment cost
• visual evidence

Innovation

• hybrid encoding
• skill evolution
• vision-language models
• multimodal agent
• self-evolving agent