The documentation contains merely one sentence on this topic:
https://docs.oracle.com/database/121/TGSQL/tgsql_optop.htm#GUID-4180BA97-3E2C-41F9-B282-4FB3FF9532CB
The BATCHED access shown in Step 1 means that the database retrieves a
few row ids from the index, and then attempts to access rows in block
order to improve the clustering and reduce the number of times that
the database must access a block.
Consider a below example of the (simplified) index
+-------------+------------------+
| index value | block nbr-rowid |
+-------------+------------------+
| 1 | 015-000123 |
| 2 | 034-000527 |
| 3 | 088-000285 |
| 4 | 015-000889 |
| 5 | 088-000632 |
........
........
In the "normal" (not batched) method Oracle retrieves rows in the order determined by the index:
- retrieves block 15, then retrieves row 015-000123 from this block
- retrieves block 34, then retrieves row 034-000527 from this block
- retrieves block 88, then retrieves row 088-000285 from this block
- retrieves block 15 (again), then retrieves row 015-000889 from this block
- retrieves block 88 (again), then retrieves row 088-000632 from this block
In the batched method oracle retrieves a few entries from the index, then first sorts them by the number of block, then process entries in the order determined by number of blocks:
- retrieves block 15, then retrieves rows 015-000123 and 015-000889 from this block
- retrieves block 34, then retrieves row 034-000527 from this block
- retrieves block 88, then retrieves rows 088-000285 and 088-000632 from this block
As you see in this example, blocks were fetched only 3 times instead of 5 times, so a number of block reads from the disk has been reduced - some blocks have been read only once instead of two (or more) times.