Communication is fundamental to cyber-physical agents. Individual agents often possess only partial observations or complementary capabilities, and therefore must exchange information to accomplish diverse cyber-physical tasks. Transmitting raw sensor observations incurs substantial communication overhead, while natural-language communication adds computational overhead from generating text at the sender and interpreting it at the receiver. These limitations make conventional communication inefficient for real-time multi-agent systems.
Problem Formulation
Following common settings in prior work, we consider cyber-physical agents that locally encode their observations and exchange latent representations over a communication link. For an observation, a local encoder produces a latent representation:
The equation for the encoded latent representation is not included in the provided text.
Each latent token has a fixed dimension, and the latent representation has a token length. Encoders may be modality-specific, for example, for image, audio, or text inputs. Within each agent, multiple downstream tasks may coexist and generate latent messages that share the same communication link.
Semantic Change Evaluation
Physical observations often evolve continuously, creating substantial redundancy between consecutive observations. This stage measures that semantic redundancy and identifies the nonredundant information to transmit. Directly comparing two latent messages token by token can be unreliable because their corresponding observations may be misaligned due to object or viewpoint motion. SCALE therefore first establishes token-level correspondences between the current latent message and the receiver’s acknowledged latent state before quantifying semantic changes.
For each current token, SCALE first finds the most similar token in the receiver’s acknowledged latent state:
The matching equation is not included in the provided text.
The corresponding maximum cosine similarity is then used to quantify the semantic change of the current token:
The semantic-change equation is not included in the provided text.
Highly similar tokens receive small change scores, while tokens with low similarity receive larger scores. A similarity threshold determines whether the matched receiver-side token can be reused; a match is reusable only when it satisfies that threshold. Matching is performed against the receiver’s acknowledged state to ensure that only information confirmed to be available at the receiver is reused.
Utility–Cost Profiling
The semantic change scores capture the amount of new information carried by individual latent tokens. This stage further characterizes the trade-off between semantic gain and communication cost at different transmission sizes, where the transmission size denotes the number of latent tokens sent for a message.
Tokens with larger semantic change scores are prioritized by sorting them in descending order. Cumulative semantic gain and communication cost are then defined:
The equations defining cumulative semantic gain and communication cost are not included in the provided text.
Benchmark Performance
Task-specific performance metrics are used to evaluate two tasks on CooperScene. For 3D object detection, the reported metrics are average precision (AP) and message on-time rate. AP is computed using an intersection-over-union threshold of 0.5 to match predicted boxes to ground-truth boxes.
For trajectory prediction, the reported metrics are task success and message on-time rate. Task success is the percentage of prediction targets for which at least one of six predicted trajectories is within 2 m of the ground-truth position.
Limitations
SCALE improves bandwidth efficiency and supports real-time scheduling for cyber-physical system agents, but it has two limitations.
First, its communication savings depend on spatiotemporal redundancy in latent representations. Observations in the physical world can change abruptly, which may reduce reusable information and increase the communication required to maintain task performance.
Second, the scheduling algorithm assesses deadline feasibility using the current link rate. It cannot handle unexpected bandwidth decreases that occur on timescales shorter than the minimum scheduling interval, which may affect message timeliness before the scheduler can adapt.
Frequently Asked Questions
How does SCALE reduce communication overhead? It prioritizes latent tokens according to semantic change and reuses matched information from the receiver’s acknowledged state when the similarity threshold is met.
Why does SCALE match tokens to the receiver’s acknowledged state? The acknowledged state confirms which information is available at the receiver, so only confirmed information can be reused.
What metrics are used for the CooperScene object detection task? The evaluation reports average precision and message on-time rate. Average precision uses an IoU threshold of 0.5 to match predicted boxes to ground truth.
What can limit SCALE’s message timeliness? Unexpected bandwidth decreases faster than the minimum scheduling interval can affect message timeliness before the scheduler adapts.