Morpho Announces VLM Solution Linking Video Recognition to Response Design

Morpho announced “Morpho VLM Fluid Action” on November 10, a solution that uses a Vision-Language Model (VLM) to recognize situations in video and design responses. response.jp reported on November 11 that the system makes an initial decision on an edge device and connects to the cloud as needed.

Morpho says the solution can recognize situations in video and configure responses without preparing training data. Linking video recognition to action design is relevant to applications that require decisions to be built individually after detection, but the information provided does not establish the conditions for its use or the safeguards in place.

The key point of the announcement is that Morpho presented a product that links recognition in video—not just identifying objects—to the design of responses. According to the company, the solution can recognize situations without prepared training data and configure actions based on the results. This offers a different starting point from the conventional approach of collecting data and building recognition systems for individual objects or scenarios. However, the information provided does not make clear how broadly the claim that no training data is needed applies across configurations and applications. In Morpho starts technology development for automotive simulators; set to debut at dSPACE event, we covered the development of technology to convert real-world driving data into scenarios for simulators. [Morpho selected for Tokyo Metropolitan Industrial Technology Research Institute’s demonstration-based joint research; to develop measurement services for indoor autonomous inspection drones](/en/news/morpho-selected-for-tiri-joint-research-to-develop-indoor-autonomous-i-2abeef) showed the deployment of image-processing and AI technologies in industrial applications. Unlike development projects for these specific applications, the new solution presents a common framework for designing video understanding and action selection to suit different uses. However, there is no confirmation that the technologies developed in those projects are incorporated into this product. The setup, in which an initial decision is made at the edge and the cloud is used as needed, points to an approach that allocates video processing between locations depending on the application. At the same time, latency, handling of communication outages and the scope of processing performed on the edge are issues that could affect whether the system is viable in real-world operations. The types of actions that can be configured based on recognition results, safeguards such as stopping or issuing alerts in the event of a misrecognition, and the conditions used to measure performance also require clarification. Going forward, key questions will include the actual use cases, the criteria for determining whether cloud connectivity is needed, and how the system will be evaluated after deployment. See the original source for details.