RI Seminar: Avik De : Design and control of insect-scale bees and dog-scale quadrupeds

In this talk

1. 2:10

   Simple models for robots at different scales

   De connects his work on insect-scale flyers, small hopping robots, and commercial quadrupeds through the limits of power, force, and computation. He uses reduced models called templates to describe desired behaviors, then anchors those behaviors in the full robot. Much of the control work he presents is reactive and focused on practical applications.

2. 7:23

   Composing behaviors and adapting to underactuation

   De describes combining simple behavior models, such as vertical hopping and lateral position control, to coordinate a robot’s movement. Later, he considers how a legged robot’s available control changes with its stance and actuation. His approach selects a suitable reduced model for the robot’s current configuration, then uses it for planning or reactive control, including leaping demonstrations on Minitaur.

3. 16:41

   Controlling RoboBee flight and designing its wings

   For RoboBee, De pairs a model predictive controller with an inverse dynamics method that maps desired forces to actuator inputs. Simulations show improved trajectory tracking over a tuned reactive controller, including perching and flips; a numerical wrench map also helps compensate for simulated actuator manufacturing bias. He then describes optimizing wing and transmission parameters, finding that a nonlinear transmission can improve performance and specific lift.

4. 41:05

   Building efficient quadrupeds for field use

   At Ghost Robotics, De says the team favors common hardware, backdrivable transmissions, and software based estimation and control, while designing robots for wet environments. The larger robot can travel about 12 kilometers per charge and carry up to 20 pounds. Current applications include security patrols and mapping construction sites and mines; future work includes manipulation and greater autonomy.

5. 50:17

   Open questions in mobile manipulation

   In the discussion, De describes demonstrations in which a quadruped stands on three legs to push or operate a doorknob, while noting that a separate arm may be more practical for customers. He suggests that coordinated bracing can reduce the force needed to open a door, and that dragging may matter more than lifting in some bomb disposal tasks. He is unsure whether locomotion’s template composition methods will transfer well, since the body and arm may need to be controlled as a coupled system.