New DPTC Framework Boosts Rocket Landing Robustness Against Saturation

Reusable launch vehicles must execute a high-angle-of-attack flip maneuver to return to a landing site, but highly nonlinear dynamics, aerodynamic uncertainties, and actuator saturation make robust trajectory optimization incredibly difficult. In a new paper on arXiv, researchers Liwei Chen and Tong Qin introduce a differentiable physics framework called Differentiable Particle Tube Control (DPTC). At its core, DPTC uses an ensemble-based distribution shaping strategy to model state uncertainty as a Lagrangian particle ensemble. Crucially, the framework embeds hard actuator projection operators directly into the computational graph, allowing end-to-end backpropagation to jointly optimize the nominal feedforward trajectory and a time-varying feedback policy. This means the controller can proactively anticipate and avoid actuator limits rather than just reacting after saturation occurs.

The team evaluated DPTC against an automatic differentiation-based Successive Convexification (AD-SCvx) baseline combined with a conventional covariance steering feedback strategy in six-degree-of-freedom Monte Carlo simulations. While the baseline nominally achieved fuel-optimal solutions, its unconstrained feedback formulation became susceptible to actuator saturation under aerodynamic disturbances, leading to degraded closed-loop robustness. In contrast, DPTC performed a constraint-aware performance trade-off: it deliberately relaxed spatial tracking accuracy to preserve critical control authority. This demonstrates that integrating differentiable physics with ensemble-based optimization provides an effective, practical framework for robust guidance in highly constrained aerospace flight systems. The approach could significantly improve the reliability of reusable rocket landings in real-world, uncertain conditions.