Soft $Q(λ)$: A multi-step off-policy method for entropy regularised reinforcement learning using eligibility traces