Robotics at Rensselaer

1
Robotics at Rensselaer

• Faculty
• Srinivas Akella
• Wes Huang
• Volkan Isler
• Jeff Trinkle
• John Wen

Thanks to Barry Bharathm from the GRASP Lab at
University of Pennsylvania for providing the last
four slides.
2
Summer 2004 at Lake George
3
Current Research Foci Motivations and Issues

• Model-based manipulation planning
• Robots can observe unstructured environs, but
  cannot do physical work
• Simulation with unilateral contact one cannot
  plan if one cannot predict
• Curse of dimensionality
• Dimension of space of inputs and design
  parameters is very large
• Need methods to prune away large chunks of the
  search space
• Family of models of mechanics is needed
• Methods must provide tractible mechanism for
  handling uncertainty
• Exact motion planning
• Exact MP algorithms for general problems are
  complete but intractable
• Sample-based algorithms suffer inefficiencies due
  to lack of knowledge of global structure of
  C-space
• We are extending the space of problems solvable
  by exact methods
• We are using knowledge of global topology to
  inform sample-based methods

4
Hierarchical Family of Models

• Models range from pure geometric to dynamic with
  contact compliance
• Required model resolution is dependent on
  design or planning task
• Approach
• Plan with low resolution model first
• Use low resolution results to speed planning with
  high resolution model
• Repeat until plan/design with required accuracy
  is achieved
• Current modeling research
• Focusing on a continuous family of models
  covering the region shaded region
• Understanding relationships will facilitate
  translating results across model upgrade steps
  during planning and design

Rigid
Compliant
Dynamic
Quasistatic
Kinematic
Geometric
5
Two Examples
Parts Feeder Design

• Parts feeder design goals
• Exit orientation independent of entering
  orientation
• High throughput
• Design geometry of feeder to guarantee 1) and
  maximize 2).
• Feeder geometry has 12 design parameters
• Evaluate feeder design via simulation

6
Geometric feasibility of part feeder design
using RRT

• RRT simulation finds Geometrically feasible
  design points
• in design space.

7

• RRT simulation shows Geometrically infeasible
  design parameters

8
RRT simulation finds kinematically feasible path
starting with geometrically feasible parameters
9
A feasible initial design (Geometric Kinematic)
used to bootstrap dynamic analysis and
optimization