One of the fundamental challenges in the design of autonomous robots is to reliably compute motion strategies in spite of significant uncertainty about sensor reliability, control errors, and unpredicatable events. The Partially Observable Markov Decision Process (POMDP) is a general and mathematically principled framework for this type of problem. Although exact solutions are computationally intractable, modern approximate POMDP solvers have made POMDP-based approaches practical for robotics tasks. However, almost all existing POMDP-based planning software suffers from at least two major issues. Firstly, most POMDP solvers require the model to be known a priori and remain constant during runtime, and secondly, quite a lot of the existing software is not very userfriendly. This paper presents the Toolkit for approximating and Adapting POMDP solutions In Real time (TAPIR), which tackles both problems. The need for a constant, fully known POMDP model is averted by implementing the recent Adaptive Belief Tree (ABT) algorithm, while user-friendliness is ensured by a welldocumented modular design, which also includes interfaces for the commonly-used Robotics Operating System (ROS) framework, and the high fidelity simulator V-REP. TAPIR can be downloaded from http://robotics. itee. uq. edu. au/tapir. To the best of our knowledge, TAPIR is the first software toolkit that directly addresses the aforementioned problems.