Hybrid systems—those that combine continuous dynamics with discrete events such as impacts—are notoriously difficult to control in real time because the timing of those events is usually unknown. A new preprint from Valentin Gaucher and Wenlong Zhang proposes a model predictive control (MPC) framework that lifts the switching instant into the decision vector, making it an explicit optimization variable alongside the continuous control inputs. This avoids the common practice of relying on a fixed time grid and lets the controller reason directly about when an event should occur.
The authors show that the resulting transcription is a smooth nonlinear program whose Karush-Kuhn-Tucker conditions are consistent with the event-time stationarity condition of the Hybrid Minimum Principle. That means first-order hybrid sensitivities are recovered implicitly through the KKT system, rather than by computing saltation matrices. Each MPC step then reduces to solving a single sparse quadratic program that jointly optimizes control inputs and impact timing.
On a benchmark of a bouncing mass landing on an oscillating platform, the method attached in half the time of fixed-grid saltation MPC, reduced cumulative control effort by 30%, and eliminated post-impact rebounds—at the cost of a 15% increase in per-cycle solve time. It also matched the cumulative effort of offline time-aware solvers to within 4%, while running at 2.4 ms per cycle compared with 281 ms for a complementarity-based formulation, a speedup the authors attribute largely to the transcription itself.