Venti is developing self-driving container trucks, aka Autonomous Prime Movers (APMs) to move shipping containers between cranes within the PSA Singapore port. The project aims to alleviate the strains in the supply chain by addressing manpower shortages and further improving operational efficiencies. The port offers unique challenges for the deployment of self-driving vehicles such as high environment change (variable stock of stacked shipping containers, moving cranes), high occlusions from large actors and structures, all-weather operational demands, tight positional tolerances for interfacing with cranes, interfacing with port specific infrastructure, specialised port traffic patterns with corresponding traffic rules, and control of long articulated vehicles (truck with trailer) under variable loading up to a maximum of 65 metric tonnes. Other problems are common to urban traffic scenarios, such as mixed traffic (human and robot) driver navigation of intersections and lane changes with travelling speeds of up to 40 kph, which demands APM perception range of 150m.
The wharf is designated as the few lanes alongside the vessels (large container transport ships), serviced by quay cranes. In this area, there is an absence of static global features, as the large quay cranes move on rails. There are multiple bi-directional lanes (to service container-facing requirements on refrigerated containers), and driving routes between the wharf and the yard (the remainder of the port) are made across an unmarked area called the back reach. The lift height of the cranes is also higher than in the yard, and the APMs must obey port rules to yield in the presence of suspended loads. Additionally, there will be many pedestrians in the wharf area to attach or remove small mechanisms called twist locks from the corners of the containers (these mechanisms are used for alignment and lashing to the vessels).
The focus of the graduate student research will be on the localisation problem for entry, exit and traversals within the wharf area of the port. This is a unique challenge due to
a) lack of static features on the global map,
b) demand for precise alignment to quay cranes for mounting/offloading containers,
c) high occlusion due to large structures in the surroundings, and
d) potential localisation jump issues when transitioning back to the yard, ie switching from a relative position tracking or dead reckoning back to global position feature matching methods.
This may involve tracking of the vehicle’s position relative to specialised port features and/or experimentation with new sensor modalities. Minor infrastructure changes can also be explored if and as needed.
