The bot -mechanical- system does not 'behave' equally in forward and reverse directions. You've got software and mechanical assembly and adjustments such that everything behaves -repeatably- going in one direction. That's very very cool.
You didn't mention if you've got four wheels, three wheels, a roller ball, or a plain skid for balance. In reverse the -entire- mechanical system works in reverse, with drastic effects. Everything changes, starting at the gearbox. There are no perfect gears. A -slight- misalignment or variation in the -finest- detail of a gear will cause 'different' behavior between the two gearboxes, when it's run forward and then in reverse. Try going in a purely straight line with no navigation/line follower, in forward and then reverse, and you'll see some differences.
Another example, with worse effects. A simple skid -trailing- the drive wheels, the skid -follows- the wheels and whatever variation/imperfection is sees in the surface it's sliding on has a slight but minimal effect - so you have good repeatability. Reverse the motion with the 'skid leading,' and any bumps/bounces/imperfections in the sliding surface are -amplified- by the skid in a sense 'controlling' the fine-tune of the direction. A grain of sand will throw it off. I'd bet that not only does the reverse motion not match the forward motion accurately, it's also never the same variation(s) twice in a row. It may seem completely random, and that's caused by the random surface.
I mentioned skid, but the same applies to ball or wheel. For example a wheel has a wheel and axle. When it's 'following' the variations in the surface its riding on causes it to lean and wobble different ways, on a micro scale, but not with great effect on the leading driving wheels. That's why it's fairly repeatable in forward motion. However in reverse motion the -slightest- variation in the surface the wheel is riding on, just a grain of sand, will cause a lean/wobble in whatever random direction and amount,and very effective steer the bot on a random direction.
The mechanical assembly is not perfectly aligned. It can't be. Worse, it can -never- be aligned close enough to track in reverse as it does in forward, because the variations of the surface are -amplified- in the reverse direction. If the alignment/mechanical oddities/balance were 'off' enough in the forward direction, that it couldn't track repeatable, that could be fudge-factored / fiddle-factored to correct. But the effects will -always- be worse in reverse.
A real world analogy. Put your arm out the window of a moving car, with fingers flat, and hand pointed backwards 'with the wind.' It's not hard to hold your arm in a straight line. Now point arm forward, and it's drastically harder to hold your arm in a straight line. That's because when the hand is trailing the arm it's just following and along for the ride. But when leading it 'steers' where the arm points.