 |
Publications |
|
|
|
|
|
|
 Karumanchi, S.B. |
|
Off-road mobility analysis from proprioceptive feedback
PhD thesis, The University of Sydney, Aug, 2010
|
Abstract
Current terrain perception modules in unmanned ground vehicles (UGVs) are focused on creating an accurate internal representation of the environment. Exteroceptive parameters such as terrain colour and terrain slope have little value if the vehicle cannot associate them with a value of cost/utility of movement. This thesis investigates the problem of using proprioceptive feedback to aid decision making for UGVs operating in unstructured environments such as off-road terrain. The aim is to derive a gross assessment of utility/cost of environmental conditions given past observations of proprioceptive feedback such as wheel slip. This problem of environmental assessment is termed as Scene Interpretation and is useful for decision making tasks such as path planning that need to account for the relative difficulty between different conditions in order to choose the best possible or a feasible path between two points. The principle contribution of this thesis is a novel problem formulation using Bayesian non-parametric learning techniques that interprets sensed environmental conditions in a data-driven manner. This learning architecture does not make strong assumptions about the state of the environment. Hence,it can generalise to complex feature representations that are required in unstructured environments. In addition,proprioceptive feedback is used to minimise input from an expert in the training process which makes data collection practical by allowing self-supervision at the lowest level. The proposed proprioceptive scene interpretation strategy is demonstrated on a 8x8 skid-steered vehicle operating in complex terrain slopes. Results are shown for two scenarios: first,giv en measurements from ranging sensors; second,given altimetry and aerial imagery. In addition,practical process models are presented and discussed that enable estimation of key proprioceptive states (wheel slip and traction coefficients) in an unscented Kalman filter implementation.
|
| Download |
PDF full text (5.87M)
|
|
Copyright Notice & Disclaimer |
|