To address the problems of non-instantaneousness and poor adaptation to grid maps in the warehouse robot Seeker Optimization Algorithm (SOA) path planning, an improved method is proposed to use the concept of key nodes to represent path critical information for iterative search during population (path) initialisation and update, followed by node connection to form a complete path after the search is completed. Within this framework, a node correction method is firstly proposed to correct the key nodes after the optimization search, and secondly a node extension method is proposed to connect the nodes, in order to address the problem of node connection. Then a collision avoidance strategy of collision detection-replanning is proposed for the problem of multi-robot path anti-collision in the storage workshop by combining a reservation table of time and position. Finally, the improved SOA path planning and the proposed collision avoidance strategy are validated through software simulation. The results show that for single path planning, the improved SOA not only greatly reduces the seeking time greatly, but also improves the path redundancy significantly; and the proposed collision avoidance strategy with the improved SOA can effectively avoid robot collision in the storage workshop.