GhostShell: Streaming LLM Function Calls for Concurrent Embodied Programming

We present GhostShell, a novel approach that leverages Large Language Models (LLMs) to enable streaming and concurrent behavioral programming for embodied systems. In contrast to conventional methods that rely on pre-scheduled action sequences or behavior trees, GhostShell drives embodied systems to act on-the-fly by issuing function calls incrementally as tokens are streamed from the LLM. GhostShell features a streaming XML function token parser, a dynamic function interface mapper, and a multi-channel scheduler that orchestrates intra-channel synchronous and inter-channel asynchronous function calls, thereby coordinating serial-parallel embodied actions across multiple robotic components as directed by the LLM. We evaluate GhostShell on our robot prototype COCO through comprehensive grounded experiments across 34 real-world interaction tasks and multiple LLMs. The results demonstrate that our approach achieves state-of-the-art Behavioral Correctness Metric of 0.85 with Claude-4 Sonnet and up to 66X faster response times compared to LLM native function calling APIs. GhostShell also proves effective in long-horizon multimodal tasks, demonstrating strong robustness and generalization.

Deep Smart Contract Intent Detection

Nowadays, security activities in smart contracts concentrate on vulnerability detection. Despite early success, we find that developers' intent to write smart contracts is a more noteworthy security concern because smart contracts with malicious intent have caused significant users' financial loss. Unfortunately, current approaches to identify the aforementioned malicious smart contracts rely on smart contract security audits, which entail huge manpower consumption and financial expenditure. To resolve this issue, we propose a novel deep learning-based approach, SmartIntentNN, to conduct automated smart contract intent detection. SmartIntentNN consists of three primary parts: a pre-trained sentence encoder to generate the contextual representations of smart contracts, a K-means clustering method to highlight intent-related representations, and a bidirectional LSTM-based (long-short term memory) multi-label classification network to predict the intents in smart contracts. To evaluate the performance of SmartIntentNN, we collect more than 40,000 real smart contracts and perform a series of comparison experiments with our selected baseline approaches. The experimental results demonstrate that SmartIntentNN outperforms all baselines by up to 0.8212 in terms of the f1-score metric.