is there any known instance of attack in practice ?
No. In cryptography, we try to stop using constructions/protocols long before they're realistic to perform.
And while no 80-bit collision attack is known to have been performed, we do have evidence of a computation that performed a multiple of the amount of work needed for one: the totality of Bitcoin's proof of work as of May 1st 2023 is estimated to be over 294 SHA256 hashes. The hardware produced and used for Bitcoin mining cannot be used for collision attacks on Bitcoin address creation, but it does show that amount of computation is theoretically within reach of humanity.
Also what addresses are valid to use, if person what to avoid 80-bit collision attacks ?
First of all, this attack is one on the creation of multiparty addresses. For example, when you're constructing a multisig address together with someone else, that other party may try to give you a key such that when combined with your key yields an address that they can spend on their own. It is not applicable to addresses which only involve a single party for spending. In fact, for such single-party addresses, 160-bit addresses are overkill even and 128-bit addresses would suffice.
For multi-party addresses, in settings where the collision attack actually applies, P2WSH and P2TR have sufficiently large (in bits) commitments to the keys/script used to make this attack infeasible.