On Feedback Speed Control for a Planar Tracking

This paper proposes and validates a novel feedback speed control law paired with constant bearing steering that ensures asymptotic stability and input-to-state stability for planar tracking in leader-follower formations, demonstrating convergence to periodic orbits and scalability to N-agent networks.

The Gist

Imagine you are walking down a busy street with a friend. You want to stay perfectly side-by-side (an "abreast" formation) so you can chat easily, even though your friend is walking in a straight line, turning corners, or speeding up and slowing down.

This paper is about teaching a robot (the "follower") how to do exactly that with another robot (the "leader"), but with a special twist: instead of just turning its wheels to stay in line, the follower changes its walking speed.

Here is the breakdown of their discovery, explained simply:

1. The Problem: The "Duck and Follow" Trap

Usually, when robots (or birds) try to follow each other, they use a "pursuit" strategy. Think of a duckling following its mother: the duckling points its nose directly at the mother and tries to catch up. If the mother turns, the duckling turns.

• The Issue: If the mother turns sharply, the duckling has to turn sharply too. If the mother speeds up, the duckling has to speed up. It's reactive and can get messy.

2. The Solution: The "Side-by-Side" Dance

The authors propose a new way to dance. Instead of chasing the leader's nose, the follower aims to stay at a fixed angle to the side (like a dance partner).

• The Secret Sauce: The follower doesn't just turn; it adjusts its speed.
  • If the leader turns left, the follower doesn't just turn left; it might speed up or slow down slightly to slide into the perfect side-by-side spot.
  • It's like a surfer adjusting their speed on a wave to stay in the "pocket" rather than just paddling harder.

3. The Two Scenarios: The "Mind Reader" vs. The "Intuitive Partner"

The paper tests this idea in two different situations:

Scenario A: The Mind Reader (Ideal World)
Imagine the follower robot has a magical telepathic link to the leader. It knows exactly when the leader is about to turn and how fast they are going.

• Result: The follower calculates the perfect speed and steering instantly. They lock into a perfect side-by-side formation and stay there forever, no matter how the leader moves. The system is mathematically proven to be stable.

Scenario B: The Intuitive Partner (Real World)
In the real world, you can't read minds. The follower robot doesn't know the leader's steering inputs. It only sees where the leader is right now.

• The Challenge: The follower has to guess. If the leader suddenly spins, the follower might wobble a bit.
• The Result: The paper proves that even without knowing the leader's mind, the follower is "Input-to-State Stable."
  • Translation: If the leader makes a sudden, crazy move, the follower might drift a little, but it won't crash or spiral out of control. It will quickly find its balance again.
  • The Periodic Bonus: If the leader starts dancing in a repeating pattern (like turning left, then right, then left, then right), the follower will eventually catch the rhythm and start dancing the exact same repeating pattern, staying perfectly in sync.

4. The Real-World Test: Robot Turtlebots

The team didn't just do math on paper; they built it. They used two small, real robots (TurtleBots).

• One robot drove around making turns and changes.
• The other robot had to stay side-by-side without knowing the first robot's future moves.
• Outcome: The second robot successfully tracked the first, staying close and side-by-side, proving the math works in the messy real world.

5. The Big Picture: The "Whip" Effect

Finally, they asked: "What if we have a whole line of robots?"

• Imagine a chain of 5 robots. Robot 1 leads, Robot 2 follows Robot 1, Robot 3 follows Robot 2, and so on.
• When the leader makes a turn, the "news" of that turn ripples down the line.
• The robots at the end of the chain have to move much faster (like the tip of a cracking whip) to keep up with the formation.
• Why it matters: This explains how information travels through a flock of birds or a school of fish. A change in direction by the leader creates a "wave" that travels through the group, and everyone adjusts their speed to stay together.

Summary

This paper teaches robots (and potentially future self-driving cars or drone swarms) a new trick: Don't just chase; adjust your speed to stay in sync. It proves that even if you don't know what your partner is planning to do next, you can still stay perfectly in step by listening to the rhythm of the movement and adjusting your own pace.

Technical Summary

1. Problem Formulation

The paper addresses a planar leader-follower tracking problem involving two nonholonomic agents (modeled as unicycles).

• Objective: The follower (Agent 2) must track a leader (Agent 1) while maintaining a specific "abreast" formation. In this formation, the agents move side-by-side with a fixed relative distance $\rho_0$ and relative bearing angles $\alpha_1 = -\pi/2$ and $\alpha_2 = \pi/2$ (Bertrand mates).
• Challenge: While steering control (heading) is well-studied in formation control, speed regulation has received less attention. The paper focuses on designing a feedback speed control law ($v_2$) for the follower, paired with a constant bearing steering law ($u_2$), to achieve stability.
• Scenarios: The study considers two cases:
  1. Full Information: The follower knows the leader's instantaneous steering ($u_1$) and speed ($v_1$).
  2. Leader-Independent: The follower does not know the leader's steering ($u_1$), treating it as an unknown input/perturbation.

2. Methodology

The authors utilize a geometric control approach based on shape dynamics (relative distance and bearing angles) rather than global coordinates.

• System Modeling:

  • Agents are modeled as planar unicycles with state $(r_i, x_i, y_i)$ and controls $(v_i, u_i)$.
  • Shape variables are defined as relative distance $\rho$ and bearing angles $\alpha_1, \alpha_2$.
  • The dynamics of these shape variables are derived to decouple the formation geometry from the global frame.

• Control Design:

  • Steering Control ($u_2$): A Constant Bearing (CB) strategy is employed. The follower steers to maintain a fixed bearing angle $\alpha_2 \to \pi/2$. This is achieved via $u_2 = -\mu_2 \cos \alpha_2 + \frac{1}{\rho}(v_1 \sin \alpha_1 + v_2 \sin \alpha_2)$. This ensures exponential convergence of $\alpha_2$ to the desired value.
  • Speed Control ($v_2$):
    • Case 1 (Known $u_1$): A feedback law is derived using a potential function $g(\rho)$ and its derivative $f(\rho)$ to stabilize both distance $\rho$ and angle $\alpha_1$. The law is:
      $$v_2 = -v_1 \sin \alpha_1 + \rho (u_1 - \mu_1 \cos \alpha_1 + v_1 f(\rho))$$
    • Case 2 (Unknown $u_1$): The term involving $u_1$ is removed from the speed law. The system is then analyzed as a non-autonomous system where $u_1(t)$ acts as an external input.

• Stability Analysis:

  • Asymptotic Stability: Proven for the case where $u_1$ is known.
  • Input-to-State Stability (ISS): Proven for the case where $u_1$ is unknown. This guarantees that the formation error remains bounded and decays if the leader's steering disturbance ceases.
  • Periodic Convergence: Proven that if the leader's steering $u_1(t)$ is periodic, the follower's shape variables converge to a periodic orbit with the same period.

3. Key Contributions

1. Novel Speed Control Law: The paper proposes a specific feedback speed control law that, when paired with constant bearing steering, achieves asymptotic stability for an abreast formation. This highlights the critical role of speed modulation in formation maintenance, distinct from pure steering control.
2. Robustness to Unknown Steering: The authors prove that the system is Input-to-State Stable (ISS) even when the follower lacks knowledge of the leader's steering input. This makes the control law practical for decentralized biological or robotic swarms where derivative estimation of neighbor states is noisy or unavailable.
3. Periodic Orbit Convergence: Theoretical proof that periodic leader maneuvers result in periodic follower trajectories, characterizing the system's behavior under continuous oscillation.
4. Scalability to N-Agent Chains: The framework is extended to an $N$-agent chain network. Simulations demonstrate how steering perturbations propagate as "whip-like" waves through the chain, mimicking information propagation in biological flocks.

4. Results

The theoretical findings were validated through both numerical simulations and physical experiments.

• Numerical Simulations (MATLAB):
  • Scenario A (Full Knowledge): The follower successfully converged to the equilibrium point $(\rho_0, -\pi/2, \pi/2)$ with smooth trajectories.
  • Scenario B (Unknown Steering): The follower maintained a bounded formation error despite the leader executing a time-varying sinusoidal steering input ($u_1 = 0.5 + \sin(\pi t)$). The shape variables converged to a periodic orbit matching the leader's frequency.
• Robotic Experiments:
  • Implemented on two TurtleBot 3 Burger robots using an OptiTrack motion capture system.
  • The follower used the leader-agnostic control law (Prop. 3.2).
  • Results confirmed that the robots successfully maintained the abreast formation and tracked the leader's periodic maneuvers, validating the ISS property in a real-world hardware setting.
• N-Agent Extension:
  • Simulations of a 5-agent chain showed that a Gaussian steering impulse applied to the leader propagated down the chain. Agents on the "outside" of the turn increased speed to maintain formation, reaching saturation limits and creating a wave-like motion, consistent with biological flocking behaviors.

5. Significance

• Biological Insight: The work provides a mathematical framework for understanding speed-curvature trade-offs observed in nature (e.g., birds, fish). It suggests that speed modulation is a primary mechanism for maintaining spatial organization in collectives, not just steering.
• Engineering Application: The proposed control laws are robust to sensor noise and lack of global information, making them suitable for decentralized multi-robot systems.
• Information Propagation: The extension to $N$-agent chains offers a model for how directional information and perturbations travel through a flock, bridging the gap between simple two-agent tracking and complex collective behavior.

In conclusion, this paper establishes that speed feedback is a necessary and powerful tool for formation control, offering rigorous stability guarantees even under conditions of incomplete information, and successfully bridging theoretical control design with biological observations and robotic implementation.