Research Interest

My interest lies in the study of voluntary actions in general and the emergence of symbolic intelligence from them. In particular, I have been studying natural voluntary arm movements in the context of reaching for and grasping an object, obstacle avoidance, the acquisition and retrieval of a motor program, and more recently on the performance of a parietal patient and of patients with Parkinson's disease. Over the years this work has uncovered a couple of intriguing geometric invariants that can begin to form an abstract vocabulary of symbols useful to compress entire families of voluntary actions sharing common goals.

I started working on this problem under the guidance of David Zipser who was my PhD thesis advisor at the Department of Cognitive Science at the University of California, San Diego. The past 5 years I have continued to do theoretical work in combination with neural and behavioral recordings at CALTECH in the laboratory of Richard Andersen. There are several results from this work that are currently being written and published.

In my dissertation, I developed a theoretical framework that bridges perception and action through a geometric step. This transition is proposed to be independent of the time varying dynamics of movement. It explains why in primates the curves described by voluntary hand movement trajectories are separable from their temporal structure.

In recent years I have addressed the validity of the space-time separability proposed by this theoretical framework and have found supporting evidence in the neural data from extra-cellular single unit recordings performed as subjects learned to acquire and retrieve a motor program. The idea is also supported by the behavioral data from these subjects, from healthy humans and from humans with a compromised system. The latter come from an ongoing and very fruitful collaboration with Howard Poizner at the Institute for Neural Computation at UCSD. Through this collaborative effort we have gained access to the strategies of a system with a Parietal lesion confined to the left lobe and to those of Parkinson's patients.

The three most intriguing aspects of these research efforts have been

(1) The discovery of two complementary classes of cells in the Reach Region of the PPC. One class has a very sparse spike response during the first few trials of learning a new motor program in naive subjects (spanning several minutes of very few spikes and then a recovery period). Those resemble the sparse code patterns reported in visual or hippocampal responses to natural stimuli. Our experimental paradigm appears to have been a type of 'natural' stimuli for these cells in the Reach Region of the PPC. The other class has the opposite response of increasing its spike frequency abruptly and then recovering from that a few seconds later. Together they can form a gating mechanism to regulate the level of self supervision (or motor awareness) when running automatic motor programs. The activity of these cells can tell even a second prior to movement whether the system will be in "auto-pilot" mode running an automatic motor program or adjusting it to some degree.

(2) The discovery that these Reach Region neurons encode the arm posture independent of the visual cue signaling the external target. These visually responsive cells encode the postural changes both when the arm is passively repositioned and when the changes are naturally evoked so that subjects do it voluntarily. In the planning stage of motion, several hundred milliseconds prior to the actual movement, the mean firing rates of these cells can unambiguously predict for each external target location the geometric transformation that will make the arm compliant with the upcoming first impulse of the hand trajectory. A subset of these cells acquiered or sharpened their spatial tuning during these learning experiments. Some expressed the well known gain-type responses but others shifted their tuning quite significantly. Their response patterns are congruent with those predicted by artificial units in a neural network that learned to encode goal-oriented motions of a 7 dimensional arm. This theoretical work was detailed in my PhD thesis but it could be of use today to those interested in controlling a robotic arm for neural prosthetics purposes. (Contact me if you are interested because it is a rather simple solution but it calls for a paradigm shift).

(3) There is a representation of time that is space dependent, location specific and task specific governing the first impulse of voluntary reaches. I propose that this is a map from internal motor delays to the extra-personal space. This map depends on a general goal-dependent notion of distance measured by the perceptual system between the current hand position and the desired target location. For each spatial location there is a stable value of the time that it takes for the hand to begin slowing down on its way to the external target. Because of its task and location specificity this quantity can be used as a label to uniquely identify and retrieve an existing motor program among competing motor memories sharing common goals under a given context. This spatial map of internal delays is causally linked to the PPC. It breaks down in a parietal patient with a lesion to the left lobe but it is 'repaired' from external cues alone independent of sensory-motor feedback availability or of cues anchored to his moving finger. This motor program selection is a step of the planning stage that is no doubt separable from the geometric transformation that provides the time-invariant pointer to the motor programs early in the PPC.

The space-dependent temporal parameter in question compensates for the system's intermittent internal delays. It fills in the temporal gap thus making feasible the comparison between the pre-planned action and the delayed real-time movement to the specified external location. In this way the system can self-supervise its actions. The stability of this temporal map of space enables the cortex to safely operate a fraction of time into the future. Because of this compensatory time-shift and the existence of motor programs and reflex-like primitives we can be autonomous. Cells in the posterior parietal cortex regulate the degree of self-supervision. That is possibly why we can multi-task and move without having to pay attention ... unless we wanted to.