An improved Hippopotamus Optimization (IHO) algorithm is proposed to address the uneven initial population distribution and susceptibility to local optima of standard Hippopotamus Optimization (HO) in time-optimal trajectory planning for robotic manipulators. The algorithm combines Latin hypercube sampling, a nonlinear variable-step strategy, and stagnation-triggered genetic mutation. Latin hypercube sampling improves the uniformity of the initial population, the nonlinear variable-step strategy balances global exploration and local exploitation, and genetic mutation is activated when the search stagnates. Quintic B-splines are used to construct the joint trajectories of a UR5e manipulator. Segment durations are optimized to minimize the total motion time subject to joint velocity and acceleration constraints. Benchmark-function tests and trajectory-planning simulations are conducted to evaluate the proposed method. Compared with standard HO, particle swarm optimization, and a genetic algorithm, IHO achieves the shortest motion time, faster convergence, and the best run-to-run stability among the tested methods while satisfying the prescribed kinematic constraints.