Terrain Identification And Autonomous Control For Tracked Mobile Robot

Advanced technologies,such as perception,decision-making,and intelligent control,are widely utilized in field missions with tracked mobile robots to meet the environmental challenges of complexity and diversity.There are four significant challenges:the modeling and application of the soil-track dynamic system,environment perception under different maneuvers,the uncertainty modeling between the theoretical model and real world,and motion control on the complex and changeable fields,respectively.This research focuses on the research of terrain identification and autonomous driving with tracked mobile robots.Modeling is the research foundation of tracked mobile robots.In this research,a kinematic model of the tracked mobile robot is established as the foundation of the model-based motion control.Combined with the Instantaneous Centers of Rotation(ICR)model,a fast-calculated steering dynamic model is proposed with parameter correction and experiment verification.Based on the multi-body dynamic model,uneven terrain generation and three-dimension system dynamics,a tracked mobile robot system with complex and changeable terrains is built,which satisfies the requirement of Hardware-in-the-loop(HIL)and provides a real-time laboratory platform for optimization and verification of controller.Essentially,tracked mobile robots move directionally under the mixed working conditions of multiple speeds,multiple maneuvers and multiple driving modes in the field.Besides,most of the identification research is under the scenario with straight-line maneuvers and a single forward speed,lacking consideration of other maneuvers.In this research,a terrain identification dataset with the dynamic response signals is collected under various terrains,including five dissimilar terrains and three kinds of similar terrains.Moreover,different maneuvers(straight-line moving and steady-state steering)and driving modes(front driving and rear driving modes)are both taken into consideration at different forward speeds.The statistical methods are utilized to extract features in the time domain,frequency domain and time-frequency domain.Furthermore,the features are evaluated by the covariance-based feature selection algorithm to obtain the superior feature combinations.The analysis with feature category and dimensions provides a basis for the identification research.Accurate terrain identification is beneficial to improve the moving of tracked mobile robots.A probabilistic neural network(PNN)is selected as an identifier for the identification under multi-speed straight-line maneuvers with dissimilar terrains.With the test and result analysis,the superior feature categories and dimensions are obtained to realize the optimal construction of PNN.Under different speeds and maneuvers on similar terrains,a multiple deep belief networks(MDBN)is applied to the terrain identification,while the effect of different forward speeds is also explored.The research is also extended to a complex condition,including forward speeds,maneuvers,driving modes,similar terrains.An ensemble learning method is come out to achieve terrain identification in this situation.Based on the terrainmechanics,the driving torque of tracked mobile robots requires a large number of model parameters,which has the limitations of high calculation amount and instability.For the torque prediction of tracked mobile robots,this article explores a mode-free data-driven approach,which is "torque reference+error compensation" with improved GPs.On the one hand,the reference value relies on a GP regression with monotonicity,taking into account the nonlinear relationship between input and output.On the other hand,the error compensation is realized by the GP with the combined kernel.Therefore,the combination of the two provides a reference for the control of the drive torque.Not only the complex and uneven terrain surface but also the changeable terrain categories should be considered,which are significant challenges to the autonomous control of tracked mobile robots.A hierarchical integrated path tracking framework is organized with two layers:the motion planning layer and the driving control layer.The motion planning layer depends on model predictive control(MPC)with the improvement by error feedback compensation.Meanwhile,torque prediction based on a model-free data-driven approach is the main component of the driving control layer.With the HIL system,the framework was tested and improved on the two scenarios,which are the same terrain and changeable terrains.The results verified the effectiveness and real-time application of this framework,effectively compensating for the uncertainty.In this research,the tracked mobile robots under complex terrains are modeled systematically,and a multi-terrain dataset is established.Based on the machine learning method,the terrain identification and driving torque prediction of the tracked mobile robots are developed,which are integrated into the hierarchical framework.With the hierarchical control framework,the autonomous control of the tracked mobile robot under complex terrain is realized,which provides a reference for the research on the perception and control of tracked mobile robots.