Notes

control, machine learning, and reinforcement learning for automotive and robotics

Three ways to combine RL and MPC, and what each actually guarantees

Reinforcement learning and model predictive control are usually framed as rivals. In practice engineers combine them, and the interesting question is not whether the combination works but exactly which promise survives it.

35 min read · September 13, 2026

2026 · reinforcement-learning mpc safe-autonomy control sac · safe-autonomy
From DDPG to SAC: a lineage of fixes

Deep Deterministic Policy Gradient, Twin Delayed DDPG, and Soft Actor-Critic are usually presented as three algorithms. They're better understood as one algorithm and two rounds of debugging.

13 min read · August 21, 2026

2026 · reinforcement-learning sac td3 ddpg actor-critic · reinforcement-learning-for-control