Exploiting Over-The-Air Consensus for Collision Avoidance and Formation Control in Multi-Agent Systems

1. Problem Statement

The paper addresses the challenge of controlling multi-agent robotic systems (specifically those with decoupled single-integrator dynamics) to achieve two simultaneous objectives:

1. Formation Control: Agents must converge to a specific geometric arrangement (defined by a centroid and relative displacement vectors).
2. Collision Avoidance: Agents must maintain a safe distance from one another to prevent collisions.

Key Challenges:

• Communication Efficiency: Traditional distributed control relies on orthogonal communication (Time/Frequency Division Multiple Access - TDMA/FDMA) where agents transmit individually. In dense networks with many agents, this consumes excessive bandwidth and energy.
• Interference: Wireless interference is traditionally viewed as a detrimental factor that degrades signal quality.
• Safety vs. Convergence: Integrating collision avoidance (often via potential fields) with consensus algorithms can lead to local minima where agents get "stuck" and fail to reach the desired formation.

2. Methodology

The authors propose a distributed control strategy that leverages Over-the-Air (OtA) Consensus to turn wireless interference into a computational advantage.

A. Communication Model: Wireless Multiple Access Channel (WMAC)

Instead of avoiding interference, the system exploits the superposition property of wireless signals.

• Mechanism: All agents simultaneously broadcast their state information. The receiver (each agent) observes a weighted sum of these signals due to channel fading.
• Normalization: To handle unknown, time-varying channel coefficients ($\xi_{ij}$), agents broadcast a known reference value (e.g., "1") alongside their state data. By dividing the received superimposed state signal by the received reference signal, agents can compute a normalized convex combination of their neighbors' states.
• Result: This allows agents to perform a consensus step (averaging) in a single transmission slot, regardless of the number of agents, provided the network topology is strongly connected.

B. Control Strategy: Jump-Flow Dynamics

The system operates as a hybrid dynamical system with continuous "flow" and discrete "jump" phases:

1. Flow Dynamics (Continuous Time):

   • Between communication updates ($t_k$ to $t_{k+1}$), agents move based on a control input $u_i(t)$.
   • Collision Avoidance: An artificial potential field is used. If an agent enters a "danger zone" (distance $< \delta_c$), a repulsive force $r_i(t)$ is applied to push it away.
   • Reference Tracking: Agents attempt to minimize the distance between their current position and a reference state $\vartheta_i$.
   • Safety Logic: If an agent is in danger of collision, its motion is modified to ensure it does not breach the minimum safety distance $\delta_s$.

2. Jump Dynamics (Discrete Time):

   • At update times $t_k$, agents broadcast their state (position or reference) via the OtA channel.
   • Upon receiving the superimposed signal, each agent updates its reference state $\vartheta_i$ to the calculated consensus value (the normalized sum of neighbors' states).
   • This update drives the system toward a common centroid.

C. Theoretical Analysis

• Graph Theory: The communication topology is modeled as a sequence of strongly connected directed graphs.
• Convergence Proof: The authors use non-negative matrix theory and Lyapunov stability analysis.
  • They prove that the normalized channel coefficients form a row-stochastic, irreducible, and primitive matrix.
  • This guarantees that the reference states $\vartheta_i(t)$ achieve asymptotic consensus (all agents agree on a common centroid).
  • A Lyapunov function is constructed to show that the system energy decreases, ensuring the agents converge to a stationary state.

3. Key Contributions

1. OtA-Based Consensus for Formation Control: The paper extends previous work on OtA consensus by integrating it into a formation control framework that explicitly handles collision avoidance.
2. Interference as a Resource: It demonstrates a practical method to utilize signal superposition (interference) to reduce communication overhead, transforming a traditional wireless drawback into a scalability advantage.
3. Hybrid Control with Safety Guarantees: The proposed "jump-flow" system integrates continuous collision avoidance (potential fields) with discrete consensus updates, ensuring agents never violate safety constraints while moving toward the formation.
4. Scalability Analysis: Theoretical and simulation results show that the number of required orthogonal channels remains constant (or grows very slowly) with the number of agents, unlike traditional methods where resources scale linearly or quadratically with agent count.

4. Simulation Results

The authors validated the approach using a simulation of 6 agents forming a hexagon and 4 agents forming a square.

• Performance:
  • The system successfully converged to the desired hexagonal formation.
  • Collision Avoidance: The minimum distance between agents was maintained at 6.18 units, well above the safety threshold ($\delta_s = 4$).
• Efficiency Comparison:
  • OtA Protocol: Required 213 total transmissions (3 values per step $\times$ 71 steps) to reach consensus.
  • Node-to-Node (Standard): Required 2,214 transmissions (10x more) to achieve slightly faster convergence.
  • Broadcast (No Interference): Required 136 transmissions but needed 2n orthogonal channels (scaling poorly with $n$).
  • Conclusion: The OtA approach offers the best trade-off, drastically reducing resource usage (transmissions and orthogonal channels) while maintaining acceptable convergence speed.
• Local Minima:
  • In a perfectly symmetric scenario (4 agents, symmetric initial positions, balanced topology), the system converged to a local minimum (agents stopped moving but didn't form the target shape).
  • However, introducing asymmetries (randomly changing network topologies or channel coefficients) allowed the system to escape local minima and reach the desired formation. This suggests the method is robust in real-world scenarios where perfect symmetry is rare.

5. Significance and Future Work

• Scalability: The primary significance is the superior scalability of the proposed method. As the number of agents increases, the communication burden of traditional methods becomes prohibitive, whereas the OtA approach remains efficient. This is critical for 6G-enabled large-scale swarm robotics.
• Practicality: The integration of collision avoidance makes the theoretical consensus protocol viable for real-world deployment where safety is paramount.
• Limitations: The paper acknowledges that in pathological cases (perfect symmetry), the system may settle in a local minimum.
• Future Directions:
  • Developing strategies to detect and escape local minima.
  • Extending the control law to agents with more complex dynamics (e.g., double integrators, non-holonomic constraints).
  • Experimental validation using physical mobile robots.

In summary, this paper presents a robust, scalable, and energy-efficient framework for multi-agent formation control that fundamentally rethinks the role of wireless interference, turning it into a mechanism for efficient distributed computation.