%0 Journal Article %1 Bullock:1993 %A Bullock, Daniel %A Grossberg, Stephen %A Guenther, Frank H %D 1993 %J Journal of Cognitive Neuroscience %K DIRECT, Forward VITE, adaptive, delta feedback, internal inverse kinematics maps, model, rule, tool use, %P 408-435 %T A Self-Organizing Neural Model of Motor Equivalent Reaching and Tool Use by a Multijoint Arm %V 5

@article{Bullock:1993, added-at = {2009-06-26T15:25:19.000+0200}, author = {Bullock, Daniel and Grossberg, Stephen and Guenther, Frank H}, biburl = {https://www.bibsonomy.org/bibtex/2e4e6a40de7a04410abda4a17f68fc1cb/butz}, comment = {The DIRECT model puts up a framework for reaching with or without (correct feedback) with and without tools. Thereby, the model computes in every timestep for each desired movement direction an action that drives the network toward the goal. Thereby, it may rely on external or internal feedback Internal or External feedback are taken into account to calculate the current end effector position in spatial coordinates. This position is substracted from the perceived goal position to calculate a difference Vector that points from the hand to the target. The Movement along the DV is the desired output. The DV is mapped onto an motor vector using a Lookup table for different direction and joint configurations (ca. 10000 cells). The motor command is transformed by the plant or fed into a forward model that calculates the next state of the arm. Learning takes placed by Training a State x Desired Direction Lookup-Table / NN onto actions using a Delta Rule. Learning data is provided by executing random actions and observing the resulting DV of each action. Since there is no amplitude in the desired direction, this works well. The same applies for the FM. The Model is able to use Tools (new arm geometry), compensate for prisms goggles, works with clamped elbow (reduced dof) and with nonlinear motor to action mappings. Problems: Learning mechanism (not to be thought as realistic): Curse of dimensionality Goal directed Learning can not allow for 1-to-many mappings no systematic evaluation of the system no dynamics included does not compensate delays remains mostly on closed loop control to guide actions}, description = {diverse cognitive systems bib}, interhash = {d41b20d10b3b2ccb594ba58acc9f6546}, intrahash = {e4e6a40de7a04410abda4a17f68fc1cb}, journal = {Journal of Cognitive Neuroscience}, keywords = {DIRECT, Forward VITE, adaptive, delta feedback, internal inverse kinematics maps, model, rule, tool use,}, owner = {martin}, pages = {408-435}, timestamp = {2009-06-26T15:25:23.000+0200}, title = {A Self-Organizing Neural Model of Motor Equivalent Reaching and Tool Use by a Multijoint Arm}, volume = 5, year = 1993 }