⚡ electrical engineering

TDMA Based Communications Control Co-Design for Cooperative Carrying: Delay Calibration and Sampling-Rate Optimization

This paper presents a physics-based simulation study demonstrating that combining adaptive sampling rate adjustment with rotating leadership in TDMA-based multi-robot cooperative carrying systems effectively balances control stability, communication efficiency, and fair airtime distribution under realistic network constraints.

Original authors: Zahra Seifaei, Maximilian Luebke, Torsten Reissland, Danial Dehghani, Norman Franchi

Published 2026-08-11

📖 7 min read🧠 Deep dive

CC BY 4.0

Original authors: Zahra Seifaei, Maximilian Luebke, Torsten Reissland, Danial Dehghani, Norman Franchi

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a group of friends trying to carry a giant, awkward sofa up a flight of stairs. If they all shout instructions at once, nobody hears anything, and the sofa crashes. If only one person shouts while the others just listen, that one person gets exhausted, and if they trip, the whole team fails. This is the daily struggle of multi-robot teams: groups of machines working together to move heavy things. In the real world, these robots talk to each other using wireless signals, which are like invisible walkie-talkie channels. But these channels aren't perfect; they get crowded (causing delays), drop messages (packet loss), and sometimes only one robot gets to talk while the others wait in line.

The big question scientists are asking is: How do we make sure these robot teams stay stable and don't drop their cargo, without wasting all their battery power shouting unnecessary updates? The answer lies in co-design, a fancy way of saying "let's design the robot's brain and its walkie-talkie together, instead of treating them as separate problems." Instead of just shouting on a fixed schedule (like a metronome), what if the robots could listen to the network, realize it's getting crowded, and slow down their shouting? And what if they took turns being the "team captain" so no single robot gets burned out? This paper explores exactly that, using a super-accurate computer simulation to see if these tricks actually work in a world that feels just like the real one.

The Robot Sofa Heist: A Story of Smart Talk and Fair Turns

In this study, researchers set up a virtual scenario where four robots are tasked with carrying a heavy, rigid box across a flat floor. Think of them as a squad of four droids lifting a giant crate. To keep the crate from wobbling or falling, the robots need to talk to each other constantly, sharing their position and speed. In a perfect world, they would just shout their status every 60 milliseconds, no matter what. But in the messy real world, the "air" they shout through gets clogged with traffic, causing delays. If a robot shouts too often when the channel is busy, the messages get lost, and the team might drop the box.

The researchers tested three different strategies to see which one kept the box safe and the robots happy. They ran these tests inside MuJoCo, a physics engine so realistic it simulates how the robots actually bump and push against the floor, not just how they think they should move.

Strategy 1: The Stubborn Shouter (Fixed Sampling)

The first team, let's call them the "Old School," used a fixed sampling rate. They shouted their status exactly every 60 milliseconds, no matter how busy the network was. They also had a static leader: Robot #1 was the captain forever, and the other three just listened.

The Result: When the network got crowded (simulating a traffic jam), the Old School team kept shouting at the same frantic pace. This clogged the channel even more, causing many messages to get lost. Their "packet loss" rate hit 17.1%, meaning nearly one out of every six messages vanished into thin air. Worse, Robot #1 did all the talking, hogging the entire "airtime" while the others sat idle. It was unfair and inefficient.

Strategy 2: The Smart Shouter (Adaptive Sampling)

The second team, the "Smart Talkers," kept Robot #1 as the permanent captain but changed how they shouted. They used a dynamic sampling strategy. Their robot brain monitored the network delay. If the channel was clear, they shouted frequently. But if they sensed the channel getting clogged (like during the simulated "congestion" phase from 60 to 120 seconds), they slowed down, stretching their shout interval to up to 120 milliseconds.

The Result: This was a game-changer. By shouting less often when it was busy, they cleared the traffic jam. Their packet loss dropped to 11.1%, a massive improvement. They proved that you don't need to shout constantly to stay safe; you just need to shout smartly.

Strategy 3: The Fair Rotating Squad (Adaptive + Rotation)

The third team, the "Fair Squad," combined the smart shouting with a new twist: rotating leadership. Every 10 seconds, they swapped the captain. Robot #1 led for 10 seconds, then Robot #2 took over, then Robot #3, and so on.

The Result: This team achieved the best of both worlds. They kept the packet loss low (around 13.3%, slightly higher than the Smart Talkers but still way better than the Old School) because they still shouted smartly. But the real win was fairness. In the Old School and Smart Talker teams, the captain used 100% of the talking time, leaving the others with 0%. In the Fair Squad, the "airtime" was split almost perfectly evenly. Their Jain's fairness index jumped from a terrible 0.25 (total monopoly) to a near-perfect 0.91. This means every robot got a fair shot at being the captain, sharing the workload and the battery drain.

The Big Reveal: Efficiency and Fairness Can Coexist

The most exciting finding of this paper is that efficiency and fairness don't have to be enemies. For a long time, people thought you had to choose: either you optimize for speed and safety (efficiency) or you make sure everyone gets a turn (fairness). This paper suggests that by using adaptive sampling (shouting only when needed) and rotating leadership (taking turns), you can get both.

The researchers found that the two strategies address different problems. The "smart shouting" fixed the network congestion and kept the messages safe. The "rotating captain" fixed the unfairness without hurting the team's ability to carry the box. In fact, the Fair Squad had the lowest overall cost of all the teams, proving that being fair didn't make them slower or less safe.

What Happened When Things Got Tough?

The researchers didn't stop at a calm day. They simulated a "harsh" environment with even more noise, delays, and message loss.

The Good News: Even in the chaos, the Fair Squad (S2) remained the best performer. The "Old School" team struggled the most, while the "Smart Talkers" and "Fair Squad" held their ground.
The Catch: The paper did find a limitation. When they tried to scale this up to a larger team (12 robots instead of 4), the fairness started to slip. With 12 robots and the same 10-second rotation, the random swapping didn't guarantee everyone got an equal turn within the test time. The fairness index dropped from 0.91 (for 4 robots) to 0.46 (for 12 robots). This suggests that for huge teams, the robots might need a stricter, more organized schedule (like a round-robin) rather than just rotating randomly.

The Bottom Line

This paper doesn't claim to have solved every problem in robot teamwork, but it offers a very practical guide for the future. It shows that if you build a robot team that listens to the network and takes turns leading, you can move heavy loads safely without burning out your batteries or leaving half the team in the dark.

In the world of Intelligent Transportation Systems—where self-driving cars might one day platoon together or robots might work in warehouses—this "co-design" approach is a blueprint for success. It tells us that the future of robotics isn't just about faster processors or stronger motors; it's about teaching robots to be polite, efficient, and fair to each other. And as the simulations show, when robots play nice, the whole team wins.

Technical Summary: TDMA-Based Communications–Control Co-Design for Cooperative Carrying

Problem Statement
Multi-robot teams performing cooperative transportation face a fundamental tension between maintaining stable control and ensuring efficient wireless communication. In real-world Intelligent Transportation Systems (ITS), wireless networks introduce latency, packet loss, and bandwidth constraints that can destabilize coordinated motion. Conventional approaches often ignore these limitations, transmitting control data at fixed rates regardless of network conditions, which leads to inefficiency during good channel states and instability during congestion. Furthermore, traditional leader-follower architectures often assign a static leader, causing an unfair monopolization of airtime resources and creating a single point of failure. While recent work has explored adaptive sampling (adjusting transmission rates based on delay) and fair Medium Access Control (MAC) scheduling separately, there is a lack of research evaluating how these mechanisms interact under realistic wireless protocols and physics-based dynamics.

Methodology
The authors propose a communications-control co-design framework evaluated through physics-based simulation in MuJoCo, incorporating realistic Time-Division Multiple Access (TDMA) modeling (including slot overhead, jitter, and queueing delays). The study focuses on a cooperative carrying task where four robots transport a rigid payload via virtual spring-damper coupling.

The framework integrates two primary design elements:

Delay-Aware Adaptive Sampling: A feed-forward plus proportional-integral (FF+PI) controller dynamically adjusts the broadcast sampling period ( $T_s$ ). It estimates one-way delay using an Exponentially Weighted Moving Average (EWMA) and adjusts $T_s$ to keep end-to-end delay within a latency budget while avoiding unnecessary transmissions.
Rotating Leadership: A strategy where the leader role and corresponding TDMA slot rotate among team members (every 10 seconds in the primary experiment) to ensure equitable airtime distribution.

The authors compare three operational scenarios against a contention-based baseline:

S0: Fixed sampling rate with static leadership (traditional approach).
S1: Dynamic sampling with static leadership.
S2: Dynamic sampling with rotating leadership (proposed co-design).
Baseline: Fixed sampling with static leadership but using contention-based (CSMA) access instead of TDMA.

Performance is evaluated using a combined objective function ( $J_{combined}$ ) aggregating delay violations, update rate costs, and communications costs (loss and delay), alongside specific metrics for force imbalance, airtime fairness (Jain's index, Gini coefficient), and formation stability.

Key Results

Efficiency and Control Performance: Adaptive sampling (S1 and S2) significantly reduces communications overhead and packet loss compared to fixed-rate transmission (S0). Specifically, S1 reduced packet loss by 35.5% relative to S0 without degrading force-tracking quality. During network congestion, the adaptive controller increased the sampling period to reduce queueing load, effectively lowering delay violation costs.
Fairness: Leader rotation (S2) dramatically improved airtime fairness. The Jain's fairness index increased from 0.25 (static leadership, the theoretical minimum for N=4) to 0.91 (rotating leadership). This improvement was achieved with a negligible impact on the combined objective (less than 2% cost increase relative to static leadership).
Co-Design Synergy: The results suggest that adaptive sampling and leader rotation address largely orthogonal dimensions of the problem. Adaptive sampling governs efficiency and delay compliance, while rotation governs fairness. The combination (S2) achieved the lowest combined objective score of all scenarios, demonstrating that efficiency and fairness can be optimized simultaneously.
Comparison to Baselines: The proposed TDMA-based co-design (S2) outperformed a representative contention-based baseline (CSMA) in reliability, efficiency, and fairness. The contention-based approach suffered from higher packet loss (19.5% vs. 13.3% for S2) and higher combined costs, highlighting the benefits of deterministic TDMA scheduling even before adaptive mechanisms are applied.
Scalability and Robustness:
- Scalability: As team size increased from 4 to 12 robots, packet loss and formation error increased, and fairness degraded (Jain's index dropped to 0.46 for N=12) under the current random reassignment rotation scheme.
- Robustness: Under harsher channel conditions (elevated jitter and loss), the relative ordering of scenarios was preserved, though absolute performance margins shrank. Adaptive sampling retained a clear reliability advantage under severe stress.

Significance and Claims
The paper claims to be among the first to jointly examine dynamic sampling, rotating leadership, and wireless protocol interactions within a physics-realistic multi-robot cooperation setting. The primary significance lies in providing practical guidance for deploying coordinated robotic teams in resource-limited environments.

The authors modestly conclude that while the two policies (sampling and access) appear to be largely composable with limited mutual interference for overall efficiency, the independence is not absolute for all metrics; specifically, fairness shows a non-trivial, non-monotonic dependence on the rotation period that warrants further investigation. The work establishes that near-ideal airtime equity is achievable with negligible impact on task performance, offering a viable design principle for networked multi-agent ITS applications where both stability and fairness are critical. The authors acknowledge limitations, including the use of i.i.d. packet loss models rather than correlated bursty loss, and the need for hardware validation and deterministic rotation schemes for larger teams.