The scale of large nuclear power plant components makes the setup and operation of manufacturing facilities producing them highly capital intensive. An alternative approach is flexible, process-to-part robotic feature machining, which has a low capital expenditure, reduces heavy lifting requirements and takes up less permanent floorspace. A barrier to adopting robotic machine tools is that they are unable to achieve the tolerances of conventional machines tools. This thesis therefore investigates the idea that, although various levels of dimensional error accumulate in robotic machined features for complex reasons, errors are predominantly systematic and can be measured using efficient, non-contact techniques and compensated in-situ with the machining process. This represents a significant gap in the literature. An initial cost benefit analysis also finds that robotic machining could allow costs and equipment payback periods to be reduced by 76% and 69%, respectively. To validate compensation feasibility, errors are characterised using a candidate robotic machine tool. This is initially done by developing a robust robotic machining performance evaluation methodology, which makes a contribution by filling a significant gaps in robotics and machine tool standards. This guides experimental work investigating positional and machining error in a varied range of robotic feature machining operations. Results confirm that robot errors, in non-cutting and cutting conditions, are highly variable but predominantly systematic and potentially correctable. Robot error is found to be subject to the greatest degree of variance according to geometry, i.e. robot position and therefore machined feature form. This work fills a gap in the literature by thoroughly quantifying how all the complex contributors to robotic machining error relate to achievable tolerances, advancing the state of the art by providing a benchmark for judging general developments against. To solve challenges, an algorithm to compensate machining trajectories with in-situ non-contact dimensional error measurements on robot machined features was developed and tested. Simulations show that the proposed solution is successful. During practical experimentation, barriers are found and quantified relating to uneven point cloud coverage and noise. This work informs further algorithm optimisation and makes a contribution to the state of the art of robotic machining error compensation by considering total dimensional error compensation, without focus on individual error contributors, by densely measuring errors across robotic machined parts directly to account for their varying nature. Measurement challenges for compensation are investigated in depth, providing insight that is not thoroughly documented in the literature. Results quantify problems with noise, point coverage and geometrical variance. Compensation algorithm performance is then optimised by investigating a novel, efficient approach that restricts error measurement to a single direction and then smooths the compensated trajectory. This produces a high quality robotic machined surface with improved tolerance, despite machining and measurement challenges. This contributes to the state of the art of robotic machining error reduction techniques, providing new understanding on machined geometry approximation using imperfect scan data. Overall, the research presented concludes that the varying levels of dimensional error, resulting from typical robotic features machining processes, can be measured using efficient, non-contact techniques and compensated in-situ with the machining process and an efficient method for doing this has been developed.