To reduce the rebound height of a tensegrity robot during landing, this paper proposes an OSTC-DP landing-rebound control strategy based on a two-critic network architecture. First, an equivalent robot model is established from the initial geometric configuration of a six-strut tensegrity robot. Next, the physical feedback of the rebound response is obtained through dynamic motion analysis. On this basis, a constrained-optimization reward function that jointly accounts for structural safety and control performance is formulated, and the OSTC-DP algorithm is employed to learn an optimal rebound-control policy. Finally, dynamic simulations are conducted in the MuJoCo physics engine, where the physical feedback of the robot before and after control is comparatively evaluated and systematically analyzed under multiple posture conditions. The results demonstrate that, while satisfying safety constraints, the proposed method substantially reduces the landing rebound height and shortens the rebound duration, thereby validating its effectiveness for landing-rebound optimization of tensegrity robots.