research

Exploit local decompositions to track unseen objects during in-hand manipulations using vision-based tactile sensors. Leverage image translation GAN models for sim-to-real transfer.

Learning observation models in factor graphs can be viewed as shaping cost functions in energy-based learning. This enables us to learn models end-to-end efficiently even with non-differentiable optimizers in the loop.

Off-the-shelf ground penetrating radars (GPR) to localize in unknown environments. Learn GPR factors in a sparse graph optimizer.

Learn vision-based tactile observation models to be integrated within a sparse graph optimizer. Track objects during planar pushing.

Leverage primal-dual methods such as Augmented Lagrangian to solve for a constrained optimization objective online.

Online occupancy map that maintains free space for planning while adapting efficiently to dynamically changing poses from SLAM.

Planning using deformable local submaps that are efficient to update using a sparse graph-based SLAM optimizer.

Phenotyping can be formulated as an optimization in the space of plant models. Apply cross-entropy to obtain most likely samples.

Learn a mapping from multi-view in-field plant images to reconstructed 3D plant units segmented into different classes.

Modular multi-robot testbed implementing evolutionary optimization for distributed boundary coverage of a light source.

Design, system ID and control for autonomous leader-follower quadrator control using GPS feedback.