Bang-Bang Evasion: Its Stochastic Optimality and a Terminal-Set-Based Implementation

arXiv:2511.21633 · eess.SY, cs.SY, math.OC · Submitted 2025-11-26 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: Today's paper: "Bang-Bang Evasion: Its Stochastic Optimality and a Terminal-Set-Based Implementation".

Dev: The gist The optimal evasion strategy always exists, and that there is at least one such policy with a bang-bang structure.

Rosa: First, who's behind it and why it matters.

Title and authors: Rosa: Let’s talk about the paper's title and authors for a second. The full title is "Bang-Bang Evasion: Its Stochastic Optimality and a Terminal-Set-Based Implementation." It really tells you what they are tackling.

Dev: They are focusing on the intersection of optimal evasion, which we know is bang-bang in deterministic settings, and moving it into the stochastic realm where information isn't perfect.

Taro: So essentially, they’re taking a classic concept from control theory and trying to make it work when you have noise and uncertainty in your inputs or state measurements.

Rosa: Right. And the authors are Liraz Mudrik and Yaakov Oshman, who are looking at this problem from the perspective of field robotics and autonomous systems.

Dev: It’s interesting because they are applying these theoretical concepts to something very practical: avoiding interception in a planar engagement against a missile guided by linear feedback.

Taro: So what do we expect them to find? They're basically trying to prove that the optimal evasive maneuver structure holds even when you introduce randomness into the system.

Rosa: The paper aims to do exactly that. They aim to extend the known optimality of bang-bang evasion maneuvers from deterministic settings directly into this realistic, stochastic evasion setting.

Dev: And they achieve this by showing that an optimal evasion strategy always exists, and crucially, that at least one of those optimal solutions is a bang-bang policy, which means the control problem becomes finite-dimensional.

Taro: Finite dimensional what does that mean for the complexity of finding the solution?

Rosa: It means they don't have to search through an infinite number of possibilities anymore; they can focus on a finite set of optimal candidates. This is a major simplification for implementation.

The paper's summary: Dev: Now, let’s look at the actual summary of "Bang-Bang Evasion: Its Stochastic Optimality and a Terminal-Set-Based Implementation." It explains the core idea behind their proposed solution.

Rosa: The core idea is that they formulate the problem in an inherently stochastic framework, dealing with imperfect information and bounded controls, rather than assuming perfect knowledge.

Dev: They use this to prove that an optimal evasion strategy exists under these conditions, even though things are messy and uncertain. That's the first big claim they make.

Taro: So if I’m a target in the engagement, I should be able to rely on a calculated strategy rather than just reacting blindly because I don't know exactly where the missile is right now?

Rosa: Exactly. The formulation accounts for that uncertainty by having the control law factor in the posterior distribution of the state, which is what you estimate based on what you know so far.

Dev: And they use this structure to propose their specific strategy, TSE, which leverages that bang-bang structure to select between extremals at each time step based on predicted terminal outcomes.

Taro: So it’s not just picking the best move right now; it's predicting what the end of the engagement will look like and using that prediction to decide which extreme maneuver is best for this moment.

Rosa: Precisely. It leverages that predictive power derived from those state estimates to make a decision that maintains optimality over time, even under uncertainty.

The paper's improvements: Taro: So they’ve laid out the results, but what are the actual improvements they suggest for this work? They aren't just reporting a result, right?

Dev: Yes. They suggest a few ways to make it even more practical and effective. One key idea is to implement the TSE strategy in closed-loop form rather than as an open-loop system.

Rosa: That’s what they mean by reducing the control space from "infinite-dimensional control space to a finite one."

Dev: They want to make it so that instead of doing a complicated search, you get just a single score comparison that determines the current acceleration command.

Taro: So it simplifies the decision-making process down to one simple check based on those terminal sets.

Rosa: And another improvement they point out is that this improved system can perform optimal evasion against proportional navigation pursuers under uncertainty by utilizing the closed-form evaluation of terminal sets, as shown in Equation (forty-seven).

Dev: That specific equation, ★ u n T = u max T sgn(S n), is what they use to evaluate those terminal sets and determine the current acceleration command.

Taro: So that means this closed-loop implementation could actually be useful for controlling systems in real time, not just theoretical analysis?

Rosa: It’s exactly what they are aiming for. The whole point is to move from complex calculations to a single score comparison that dictates the current acceleration command in real time.

Conclusion: Dev: So to wrap things up, this paper on "Bang-Bang Evasion: Its Stochastic Optimality and a Terminal-Set-Based Implementation" shows that we can find an optimal evasion strategy even when information is imperfect.

Rosa: The main implication is that the TSE strategy gives us a concrete method for achieving superior evasion performance in these scenarios by leveraging the bang-bang structure of the problem.

Taro: What does this mean for someone who only listens to the show? It means we can expect significantly better miss distances compared to traditional methods when dealing with uncertainty.

Dev: The numbers back up that: Monte Carlo simulations showed TSE achieved a final miss distance of two point five five meters, which is significantly outperforming RTS at one point seven five meters, Singer at zero point four meters, and weaving at zero point five one meters.

Rosa: That’s a big gap in performance, and it shows that this approach generates the largest miss distances across all tested quantiles in their comparison to other strategies.

Taro: So we’re looking at a method that is computationally light enough for real-time use, with strong theoretical backing and good practical results when uncertainty is involved.

Dev: It does have limitations, though they do flag that the TSE law hedges against intercept time uncertainty only in the mean-square sense as implied by the cost function.

Rosa: So a promising direction for future work is characterizing how sensitive this TSE law is to misspecification of the intercept-time distribution, which is where we can push this further.

Taro: That sounds like a good next step: testing the sensitivity of this strategy to different assumptions about the underlying engagement timing.

Dev: It’s a solid piece of work that takes deterministic theory and makes it work for uncertain environments by providing a practical, computationally efficient framework for evasion.

Technion—Israel Institute of Technology

eess.SY, cs.SY, math.OC

Submitted: 2025-11-26

Updated: 2026-10-14

Comments: Published in the Journal of Guidance, Control, and Dynamics. 31 pages, 4 figures

DOI: 10.2514/1.G009788

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 86/100

The gist: The gist The optimal evasion strategy always exists, and that there is at least one such policy with a bang-bang structure.

Key concepts

Bang-Bang Control: This refers to a control strategy where the system always operates at its maximum or minimum allowable control input. In this context, it means the evader aggressively uses its full available maneuver capability—either maximally evasive or minimally evasive—to maximize its chances of avoiding interception.
Terminal-Set-Based Evasion (TSE): This is a specific evasion policy that uses the bang-bang structure. At each time step, it chooses between the two extremal control inputs based on predictions derived from the current state estimates and their predicted terminal outcomes, effectively hedging against uncertainty.
Stochastic Framework: The problem is modeled using probability distributions rather than fixed values. This accounts for imperfect information about the target's exact location and the missile's guidance errors, allowing for a more realistic simulation of a real-world engagement.

Summary

The gist The optimal evasion strategy always exists, and that there is at least one such policy with a bang-bang structure.

Problem Formulation

The work addresses the problem of optimal evasion in a planar endgame engagement where a target with bounded lateral acceleration seeks to avoid interception by a missile guided by a linear feedback law. The formulation is done within an inherently stochastic framework involving imperfect information and bounded controls.

The problem is mathematically formulated in a linearized, pursuer-evader stochastic engagement scenario. The state vector is defined as x = xi xi¤ q⊤ M q⊤ T⊤ ∈ R nx. The discrete-time equations of motion (EOM) are given by xk+1 = F k(eta)x k + g kM(eta)u kM + g kT u kT + omega k.

Optimal Evasion Structure

The first and main contribution is a general existence result: we prove that an optimal evasion policy exists, and that there is at least one such policy with a bang-bang structure. This extends the well-known deterministic result to the stochastic setting and establishes that the structure of optimal evasive maneuvers remains extremal even under uncertainty.

The proof relies on established findings from convex optimization theory [25, 26] and Theorem III.2. The underlying constraint set is shown to be convex and compact. Furthermore, the objective function of the evasion problem is both convex and continuous over its constraint set.

Terminal-Set-Based Evasion (TSE)

The second contribution is a specific evasion policy based on the notion of terminal sets. This policy leverages the bang-bang structure to select between the extremals at each time step based on predicted terminal outcomes derived from state estimates.

The resulting evasion law is defined as u n T = u max T sgn(S n), where sgn(·) denotes the signum function. This proposed evasion law is termed the terminal-set-based evasion strategy, and use the shorthand notation TSE in the ensuing discussion.

Performance and Comparison

Monte Carlo simulations show that the TSE strategy outperforms traditional stochastic evasion strategies based on random telegraph, Singer, and weaving models. The TSE yields the highest mean and median final miss, along with the widest spread across the 5th to 95th percentiles.

In comparison, the TSE consistently dominates RTS, producing larger evasive separations for all quantiles. The two bang-bang profiles, TSE and RTS, clearly outperform the stochastic smooth maneuvers (Singer and weaving), achieving higher miss distances across nearly the entire distribution.

Computational Efficiency

The terminal-set framework is lightweight. Constructing the short-horizon transition products and input-to-terminal gains scales as O(p max N f H̄) for the mean terms and O(p max N f H̄ 2) for the covariance matrices. This collapses the exponential search over future bang-sequences to near-linear scaling in the horizon, achieving orders-of-magnitude savings that make real-time stochastic evasion practical.

Conclusion

This work addresses the problem of optimal evasion from an interceptor guided by a linear feedback law in the presence of uncertainty and bounded controls. The TSE strategy demonstrates its effectiveness by generating the largest miss distances.

The present formulation assumes a known terminal-time PMF p f, and the TSE law hedges against intercept-time uncertainty only in the mean-square sense as implied by the cost. A promising direction for future work is to characterize the sensitivity of the TSE law, and of the RTS, Singer, and weaving benchmarks, to misspecification of the intercept-time distribution.

The paper addresses that gap by studying the optimal evasion problem for a bounded-control evader operating under uncertainty about the engagement’s state. The first and main contribution is a general existence result: we prove that an optimal evasion policy exists, and that there is at least one such policy with a bang-bang structure. This extends the well-known deterministic result to the stochastic setting and establishes that the structure of optimal evasive maneuvers remains extremal even under uncertainty. The second contribution is a specific evasion policy based on the notion of terminal sets. This policy leverages the bang-bang structure to select between the extremals at each time step based on predicted terminal outcomes derived from state estimates. The remainder of this paper is organized as follows.

The work addresses the problem of optimal evasion in a planar endgame engagement where a target with bounded lateral acceleration seeks to avoid interception by a missile guided by a linear feedback law. The control law factors in the posterior distribution of the state. We extend the well-known optimality of bang-bang evasion maneuvers in deterministic settings to the realm of realistic, stochastic evasion scenarios. The Monte Carlo simulations show that the TSE strategy outperforms traditional stochastic evasion strategies based on random telegraph, Singer, and weaving models. This work addresses that gap by studying the optimal evasion problem for a bounded-control evader operating under uncertainty about the engagement’s state.

Improvements for AI systems

Bold evasion strategy based on terminal sets (TSE) can be implemented in closed-loop form to reduce computational complexity from infinite-dimensional control space to a finite one, making real-time stochastic evasion practical by collapsing an exponential search into a single score comparison that determines the current acceleration command.
The improved system can perform optimal evasion against proportional navigation pursuers under uncertainty by utilizing the closed-form evaluation of terminal sets, as shown in Equation (47): u n T = u max T sgn(S n).
The AI system can achieve superior evasion performance compared to classical stochastic models; for instance, the Monte Carlo study showed that TSE achieved a final miss distance of 2.55 m, significantly outperforming RTS (1.75 m), Singer (0.4 m), and weaving (0.51 m).
The system can incorporate information about the interceptor's state estimation uncertainty by modeling the target's knowledge of the interceptor as the target needs to estimate it instead, allowing for a more realistic adversarial model in pursuit of evasion.
The system can utilize a terminal-set representation where "For each index i in the support of the random terminal index f, that is, each candidate terminal step with p f(i) > 0, and for each admissible guidance-law mode j ∈ [1, p max], unrolling the discrete-time closed-loop dynamics yields an affine dependence of the terminal state on the current command."

Listen

Transcript

Key concepts

Summary

Improvements for AI systems

Related papers