Frontier research directions in visual slam

I have a robot chassis with imu and a stm32 chip, the stm32 chip is connected to a jeston orin NX, the ros noetic system is preinstalled in the jeston, and I have an intel realsense d455, and I’ve already been running orb-slam2 algorithms on this set of equipment. Now I want to do visual slam related research, is there any cutting edge slam direction I can try?