Biped robots have received much attention nowadays, as they might be used in the environment especially designed for human beings. In the present thesis, gait generation problems of a 7-DOF two-legged robot on various terrains (such as, staircase, sloping surface and ditch surface) have been solved utilizing the principles of soft computing. In analytical approach, the gaits of the lower limbs have been determined using the principle of inverse kinematics and trunk motion is generated based on the concept of static balance. The generated gait is then verified for its dynamic balance after determining the position of zero moment point. Two modules of feed-forward neural networks and fuzzy logic system each are used as the gait planners. A genetic algorithm has been utilized to optimize the weights and bias values of neural network and knowledge base of the fuzzy logic system. The gait generation problems have been formulated as both unconstrained as well as constrained optimization problem and solved. In the above work, a forward method of solving the dynamics, that is, first determining the trunk and swing foot motions utilizing either neural network- or fuzzy logic-based gait planner and then identifying the torques to produce the joint motion suggested by the planner, had been developed. Sometimes, it could be difficult to control the robot, as the motor might fail to supply the required torque for obtaining joint angle variation. On the other hand, the gait planner designed based on the capacity of the motors could successfully perform the task of controlling a robot in real-time. An attempt has been made to develop an inverse dynamics learned neural network-based gait planner for the biped robot to tackle various situations, while moving through the above terrains. Here, the lower limbs' gait are determined using the concept of inverse kinematics and the trunk’s and swing foot’s motions have been derived from the inverse dynamics learned neural network. The performances of the developed approaches have been tested through computer simulations. Soft computing-based approaches are able to generate dynamically balanced adaptive gaits for the biped robot. Keywords: Biped robot, Adaptive gait generation, Staircase, Sloping surface, Ditch, Zero movement point, Dynamic balance margin, Neural network, Fuzzy logic, Genetic Algorithm.