Multi-agent trajectory planning