Multi-agent consensus problem in an obstacle-laden environment is addressed in this study. A novel optimal control approach is proposed for the multi-agent system to reach consensus as well as avoid obstacles with a reasonable control effort. An innovative nonquadratic penalty function is constructed to achieve obstacle avoidance capability from an inverse optimal control perspective. The asymptotic stability and optimality of the consensus algorithm are proven. In addition, the optimal control law only requires local information from the communication topology to guarantee the proposed behaviour, rather than all agents' information. The consensus and obstacle avoidance are validated through various simulations.