||Due to wind and wave actions, ship impacts, high-speed vehicles, and other sources of loading, structures such as offshore platforms, tall buildings, high-rise bridges, and electric transmission towers undergo significant torsional loads. Insufficient design of the foundations of these structures against torsional loads may affect the serviceability and safety of these structures. The objectives of this research are: (1) through a series of centrifuge model tests, to build basic understanding of load sharing mechanisms in pile groups subjected to torsion, pile-soil-pile interactions in torsionally loaded pile groups, and the load-deformation coupling effects in the pile groups and (2) to develop a model to analyze the problems in a rational manner. A series of centrifuge tests, including eight laterally loaded single pile tests, twelve torsionally loaded single pile tests and seven torsionally loaded pile group tests, were performed to achieve the objectives. The laterally loaded single pile tests and the torsionally loaded single pile tests were performed using a robotic manipulator. It is perhaps the first time that a robotic manipulator was used to perform the two types of single pile tests in centrifuge. The robotic manipulator was utilized to conduct multiple activities in each test and multiple tests at different loading rates. A separate testing system was developed to conduct the torsionally loaded pile group tests. The testing system can achieve pile jacking and torsional loading without stopping the centrifuge. Using this testing system, seven model tests on 1x2, 2x2, and 3x3 pile groups subjected to torsion were conducted. These tests produced the first batch of experimental data on performance of torsionally loaded pile groups. Test results of the pile groups were compared with those of the single piles. The test results reveal the load sharing mechanism. An applied torque on a pile group is primarily resisted by the torsional and lateral resistances of the individual piles in the group. The torsional and lateral resistances of the piles significantly rely on their locations in the group. The different lateral displacements of the piles at different locations are proportional to the distances from the piles to the torsional centre of the group. Consequently, different shear resistances are mobilized in the piles. Through the deflection-torsion coupling effect, the different lateral resistances then result in the different torsional resistances. In addition, pile-soil-pile interactions also influence the load sharing. Test results show that pile-soil-pile interactions are different for the piles at different locations in the pile group. The pile group test results reveal that effects of lateral movement of the pile on lateral and torsional behaviours of adjacent piles exist in torsionally loaded pile groups. In the 3x3 pile group tests, both the pile-head shear forces and the p-y curves demonstrate that side piles and the corner piles in the group interact each other. The induced reductions of the soil resistances in these piles were different. Meanwhile, the side piles were found to affect the torque distributions of the corner piles, as well as their torsional resistances. Patterns of shear zones of piles in pile groups subjected to torsion and lateral loading were found to be significantly different. The difference implies that empirical interaction factors from numerous laterally loaded pile group tests can hardly be used to analyze torsionally loaded pile groups. The deflection-torsion coupling effect was also analyzed. Generally, for a pile in a torsionally loaded pile group, the increase of the soil pressures in front of the pile in lateral loading direction contributes to increase normal stresses on the pile surface. Therefore, the torsional shaft friction increases with the group torsional angle. A factor is introduced to quantify the coupling effect in the pile group tests. A relation between the factor and lateral soil reaction is proposed. Through studying the results of the centrifuge tests, a hybrid model was developed to formulate the problem. In the model, load-transfer curves were used to model nonlinear soil behaviour, which occurs in the near fields around piles. The pile-soil-pile interactions were predicted through Mindlin's solutions. A coupling coefficient was introduced to formulate the deflection-torsion coupling effect in pile groups. This model achieved fairly good agreement with the centrifuge test results.