Hydraulically-driven parallel robots are extensively deployed in heavy machinery for their superior force-to-weight ratios, yet precise trajectory tracking remains constrained by unmeasurable internal states, strong inter-actuator couplings, and time-varying operational uncertainties. Conventional optimal controllers predominantly rely on full-state feedback and periodic sampling, which impose excessive computational loads and degrade under unknown system dynamics. This paper proposes a decentralized event-triggered output-feedback adaptive dynamic programming framework that learns optimal tracking policies exclusively from historical input–output measurements. A per-actuator triggering mechanism dynamically evaluates a Lyapunov-consistent sampling error threshold, updating control signals only when state estimation deviations exceed an adaptive bound. Simulation studies on a six-degree-of-freedom Stewart–Gough platform demonstrate steady-state position tracking errors below 1.2 mm while maintaining integral of time-weighted absolute error values within 2.5% of the periodic baseline. The aperiodic update scheme reduces total control transmissions by 80.8%, and comprehensive robustness analysis confirms uniform ultimate boundedness under parametric perturbations. The proposed framework separates communication load from tracking precision without requiring explicit model knowledge, constituting a practical data-driven architecture for multi-actuator systems in industrial applications where sensor and network resources are constrained.