Recursion à la auxiliary function
You can define this using a simple auxiliary procedure and a couple state variables – the following f implementation evolves a linear iterative process
def f (n):
def aux (n, a, b):
if n == 0:
return a
else:
return aux (n - 1, b, 0.5 * (a + b))
return aux(n, 0, 1)
print([f(x) for x in range(10)])
# [0, 1, 0.5, 0.75, 0.625, 0.6875, 0.65625, 0.671875, 0.6640625, 0.66796875]
Going generic
Or you can genericize the entire process in what I'll call fibx
from functools import reduce
def fibx (op, seed, n):
[x,*xs] = seed
if n == 0:
return x
else:
return fibx(op, xs + [reduce(op, xs, x)], n - 1)
Now we could implement (eg) fib using fibx
from operator import add
def fib (n):
return fibx (add, [0,1], n)
print([fib(x) for x in range(10)])
# [0, 1, 1, 2, 3, 5, 8, 13, 21, 34]
Or we can implement your f using fibx and a custom operator
def f (n):
return fibx (lambda a,b: 0.5 * (a + b), [0, 1], n)
print([f(x) for x in range(10)])
# [0, 1, 0.5, 0.75, 0.625, 0.6875, 0.65625, 0.671875, 0.6640625, 0.66796875]
Wasted computations
Some answers here are recursing with (eg) 0.5 * (f(n-1) + f(n-2)) which duplicates heaps of work. n values around 40 take astronomically longer (minutes compared to milliseconds) to compute than the methods I've described here.
Look at the tree recursion fib(5) in this example: see how fib(3) and fib(2) are repeated several times? This is owed to a naïve implementation of the fib program. In this particular case, we can easily avoid this duplicated work using the auxiliary looping function (as demonstrated in my answer) or using memoisation (described in another answer)
Tree recursion like this results in O(n2), whereas linear iterative recursion in my answer is O(n)

Generating a sequence for n
Another answer provided by @MadPhysicist generates a sequence for a single n input value – ie, f(9) will generate a list of the first 10 values. However, the implementation is simultaneously complex and naïve and wastes heaps of computations due to the same f(n-1), andf(n-2)` calls.
A tiny variation on our initial approach can generate the same sequence in a fraction of the time – f(40) using my code will take a fraction of a second whereas these bad tree recursion answers would take upwards of 2 minutes
(Changes in bold)
def f (n):
def aux (n, acc, a, b):
if n == 0:
return acc + [a]
else:
return aux (n - 1, acc + [a], b, 0.5 * (a + b))
return aux(n, [], 0, 1)
print(f(9))
# [0, 1, 0.5, 0.75, 0.625, 0.6875, 0.65625, 0.671875, 0.6640625, 0.66796875]
return f(n-2)//f(n-1)?? Is that the way to calculate an average?