This paper addresses the problem of robust transmission of sensed data through a vast field of small and vulnerable sensors towards a sink node. It introduces a routing algorithm called P-GRAB relying on a probabilistic gradient broadcasting framework. Our aim is to improve the GRAB algorithm by accounting for the energy expenditure and the potential of a node for creating interference in the forwarding decision of the algorithm. It is the forwarding stage of P-GRAB that differs from GRAB: once a node has the proper cost for broadcasting a packet, it decides to forward it with a given probability depending on its remaining energy level and its interference potential. We show by simulations that P-GRAB outperforms the GRAB algorithm by providing similar robustness but with much fewer forwarding packets and latency for packet delivery.